i . .

i '*
"

N A S AC O N T R A C T O R NASA CR-1"
REPORT

h
N

f L M N C13PY: RETURN TC$

AFWL (WLIL-2) !
KIRTLAND AFB, N MEX ';

PRACTICALRELIABILITY
Volume I1 - Computation
. ,"

Prepared by
RESEARCH TRIANGLZ INSTITUTE .. , ,

. _ . .

Research Triangle Park, N. C.

for

di
N A T I O N AALE R O N A U T I CASNSDP A CAED M I N I S T R A T I O N W A S H I N G T O N , 0. C. AUGUST 1968 x
i
a& . -.
 "~

. . %

TECH LIBRARY KAFB, NY

00b03bb

NASA CR-1127

PRACTICAL RE LIABILITY

Volume I1 - Computation

Distribution of this report is provided in the interest of

informationexchange.Responsibility for thecontents
resides in the author or organization that prepared it.

Prepared under ContractNo. NASw- 1448 by

RESEARCH TRIANGLE INSTITUTE
Research Triangle Park, N.C.
for
NATIONAL AERONAUTICS AND SPACE ADMINISTRATION
For sale by the Clearinghouse for Federal Scientific and Technical Information
-
Springfield, Virginia 22151 CFSTI price $3.00
 FOREWORD

. The t y p i c a l few-of-a-kind n a t u r e of NASA systems has made r e l i a b i l i t y a premium

even on t h e i n i t i a l items d e l i v e r e d i n a program. Reliabilitydefinedandtreated
on the basis of percentage of items o p e r a t i n g s u c c e s s f u l l y h a s much less meaning
t h a n when l a r g e r s a m p l e s i z e s are a v a i l a b l e as i n m i l i t a r y and commerical products.
R e l i a b i l i t y t h u s becomes based moreon engineering confidence that the item w i l l work
as intended. The key t o r e l i a b i l i t y i s t h u s good e n g i n e e r i n g - - d e s i g n i n g r e l i a b i l i t y
into the system and engineering to prevent degradation of the designed-in reliability
from f a b r i c a t i o n , t e s t i n g a n d o p e r a t i o n .
The PRACTICAL RELIABILITY series of r e p o r t s is addressed to t h e t y p i c a l e n g i n e e r
to aid his c o m p r e h e n s i o no fp r a c t i c a lp r o b l e m si ne n g i n e e r i n gf o rr e l i a b i l i t y .I n
these reports the intent is t o p r e s e n t f u n d a m e n t a l c o n c e p t s o n a particular subject
in an interesting, mainly narrative formand make t h e r e a d e r a w a r e o f p r a c t i c a l
problems i n applying them. There is l i t t l e emphasis on d e s c r i b i n gp r o c e d u r e s and
how t o implementthem.Thus t h e r e i s l i b e r a l useofreferencesforbothbackground
t h e o r y andcookbook procedures. The p r e s e n tc o v e r a g e i s l i m i t e dt of i v es u b j e c ta r e a s :
Vol.
" -~ I. - Parameter Variation Analysis describes the techniques for treating
theeffect of systemparameters on p e r f o r m a n c e , r e l i a b i l i t y , and o t h e r f i g u r e s -
of-merit .
Vol. 11. - C o m p u t a t i o nc o n s i d e r st h ed i g i t a lc o m p u t e ra n dw h e r e and how i t can
b e used t o a i d v a r i o u s r e l i a b i l i t y t a s k s .
Vol. 111. - Testingdescribesthebasicapproachestotestingandemphasizes
the practical considerations and t h e a p p l i c a t i o n s t o r e l i a b i l i t y .
Vol. I V . - Predictionpresentsmathematicalmethodsandanalysisapproaches
for reliability prediction and includes some methods not generally covered
in texts andhandbooks.
Vol. V. - Parts r e v i e w s t h e p r o c e s s e s a n d p r o c e d u r e s r e q u i r e d t o o b t a i n a n d
applypartswhich w i l l perform their functions adequately.
These reports were prepared by t h e R e s e a r c h T r i a n g l e I n s t i t u t e , R e s e a r c h T r i a n g l e
Park,NorthCarolina 27709 under NASA Contract NASw-1448. The c o n t r a c t w a s adminis-
tered under the technical direction of t h e O f f i c e of R e l i a b i l i t y a n d Q u a l i t y
Assurance, NASA Headquarters,Washington, D. C. 20546 w i t h D r . John E . Condon,
D i r e c t o r , as t e c h n i c a lc o n t r a c tm o n i t o r . The c o n t r a c t e f f o r t w a s performed j o i n t l y
by p e r s o n n e l from b o t h t h e S t a t i s t i c s Research and the Engineering and Environmental
S c i e n c e sD i v i s i o n s . D r . R. M. Burger was t e c h n i c a ld i r e c t o rw i t h W. S. Thompson
s e r v i n g as p r o j e c t l e a d e r .

iii
 This report is Vol. I1 - Computation. It serves in a support role to the other
I - Parameter Variation Analysis and Vol.
volumes, particularly to Vol. IV - Predic-
tion, by treating the computer techniques for implementing the reliability tasks
R. L. Beadles is the principal author
developed in the other volumes. of this report.
A. C. 2 and 8 ; he and J. R. Batts wrote the
Nelson made major contributions to Secs.
7.
computer programs discussed in Sec.

iv
 ABSTRACT

Thisreportplacesinperspectivetherole of a u t o m a t i c d i g i t a l c o m p u t a t i o n s i n
d e s i g nf o rr e l i a b i l i t y . It is i n t e n d e df o rt h ed e s i g ne n g i n e e r ,t h es y s t e m se n g i n e e r ,
andthe t e s t e n g i n e e r as well as t h e r e l i a b i l i t y s p e c i a l i s t . The d e g r e e o f d e t a i l
withwhichthevarioustopics are t r e a t e d i s s u f f i c i e n t t o e n a b l e t h e e n g i n e e r n o t
previously familiar with the subject to properly select and use the methods
presented.
As a f u n d a m e n t a l i n t r o d u c t i o n t o a u t o m a t i c d i g i t a l c o m p u t a t i o n , t h e r e p o r t f i r s t
brieflydescribesthecomputer, how i t i s used,and some of themathematicalproblem
t y p e st h a t are amenable t o c o m p u t e rs o l u t i o n . The o r i e n t a t i o n t o r e l i a b i l i t y is t h e n
providedin a brief perspective of reliability tasks and t h e r e l a t i o n of t h e computer
t o them. Later s e c t i o n so ft h er e p o r t treat s p e c i f i cr e l i a b i l i t yt a s k s and e x p l o r e
themathematicalmethodsrelatedto themand how t h e computer is used t o implement
them. Some s p e c i f i c computerprograms a r e i d e n t i f i e d and t h e i r u s e s i l l u s t r a t e d by
examples. Parameter v a r i a t i o n a n a l y s i s and r e l i a b i l i t y p r e d i c t i o n are t r e a t e d i n
more d e t a i l t h a n o t h e r s s i n c e t h e s e areas of a p p l i c a t i o n are p a r t i c u l a r l y s u i t e d t o
computermethods. The l a s t s e c t i o n of t h er e p o r ts u m m a r i l yt r e a t s some r e c e n t
developments i n communicatingwiththecomputerwhich make i t more s u i t a b l e t o
e n g i n e e r i n g and r e l i a b i l i t y a p p l i c a t i o n s .

V
 TABLE OF CONTENTS
Pane
FOREWORD iii
ABSTRACT V
1. Introduction 1
2. Fundamentals of Digital Computation 3
2.1 Digital Computer Concepts 3
2.2 Computer Programming Languages 7
2.3 Basic Mathematical Problems that Can Be Solved
a Computer
by 9
3. Reliability and the Computer--A Perspective 23
4. Parameter Variation Analysis 28
4.1 PVA Modeling 28
4.2 Analysis Techniques 30
4.2.1 Worst-case Analysis 30
4.2.2 Sensitivity Analysis 33
4.2.3 Moments Analysis 34
4.2.4 The Convolution Method 36
4.2.5 Monte Carlo Analysis 37
4.3 PVA Computer Programs 40
4.3.1 A General PVA Program 42
4.3.1.1 General PVA Program Example 49
4.3.2 ECAP and NASAP for PVA 58
5. Part Application Analysis 61
6. Failure Modes and Effects Analysis (FMEA) 62
7. Reliability Prediction 66
7.1 Developing the Prediction Model 67
7.2 Making the Reliability Prediction 68
7 . 3 Reliability Prediction Programs 70
7.3.1 A Computer Program for System Reliability 70
7.3.2 Reliability Cost Trade-Of f Analysis Program 84
8. Testing 94
8.1 Attribute Data 94
8.2 Variables Data 98
8.2.1 Failure-Time Data 98
8.2.2 Performance Measurements at Discrete Time(s) 99
8.2.3 Continuous Recording of Performance Measurements 101
8.3 Stress-Strength Measurements 102

vi i

d
TABLE OF CONTENTS (CONTD)
Pap;e
9. Trends i n Digital Computation 105
Appendix A 109
Appendix B 119
Appendix C 125

viii
1. Introduction
The d i g i t a l computerhashad a significant impact on engineering design and develop-
ment.Becauseof i t , l a r g e r andmore s o p h i s t i c a t e ds y s t e m sh a v e become realities
ratherthan mere dreams.But w i t ht h e s ed e v e l o p m e n t s ,t h ea c h i e v e m e n t of system re-
liabilityhas become more d i f f i c u l t . The d e s i g n e r ' s t a s k o f b u i l d i n g i n t h e relia-
b i l i t y i s a complexone.involving extensiveanalysis and computation, and i t i s o n l y
natural that the computerbeemployed t o its full capacity here also.
A good, r e l i a b l e d e s i g n r e s u l t s f r o m a continual assessment and improvement process.
P e r f o r m a n c ea n a l y s i s ,t e s t i n g ,f a i l u r e mode and e f f e c t s a n a l y s i s , and r e l i a b i l i t y p r e -
d i c t i o n are t y p i c a l , key t a s k s i n t h i s i t e r a t i v e p r o c e s s . A s a t o o l o ft h ed e s i g n e r ,
t h e computermust contributedirectlytoperformanceofsuchtasks.
Thepurpose of t h i s r e p o r t is to place in proper perspective the role of a u t o m a t i c
d i g i t a lc o m p u t a t i o n si nd e s i g nf o rr e l i a b i l i t y . It is i n t e n d e df o rt h ed e s i g ne n g i n e e r ,
t h es y s t e m se n g i n e e r , and t h e test e n g i n e e r as w e l l as t h e r e l i a b i l i t y s p e c i a l i s t . The
d e g r e eo fd e t a i lw i t hw h i c ht h ev a r i o u st o p i c s are t r e a t e d i s s u f f i c i e n t f o r e n a b l i n g
theengineernotpreviouslyfamiliarwiththesubjecttoproperlyselect and u s e t h e
methodspresented.
Of e q u a l i m p o r t a n c e t o a n a p p r e c i a t i o n f o r w h a t t h e d i g i t a l c o m p u t e r c a n do i s
a na d e q u a t ea p p r e c i a t i o nf o r what i t cannotdo.Consequently,care is taken a t
appropriatepointstoindicatethelimitations of t h e a v a i l a b l e c o m p u t e r m e t h o d s and
programs.
A s a f u n d a m e n t a li n t r o d u c t i o nt oa u t o m a t i cd i g i t a lc o m p u t a t i o n ,S e c . 2 briefly
d e s c r i b e st h ec o m p u t e r , how i t i s used,and some ofthemathematicalproblemtypes
t h a t are s o common i n many u s e s of thecomputer. The o r i e n t a t i o n s p e c i f i c a l l y t o
designreliabilityapplications i s provided i n Sec. 3 which g i v e s a b r i e f o v e r a l l
p e r s p e c t i v eo ft h ee n g i n e e r i n gt a s k s and r e l a t e s t h e r o l e of t h e computer t o them.
S e c s . 4 through 8 s e p a r a t e l y t r e a t s p e c i f i c d e s i g n t a s k s and e x p l o r e i n more
depththemathematicalmethodsand how t h e computer is usedtoimplement them. Some
s p e c i f i c computerprograms are i d e n t i f i e d and t h e i r u s e s i l l u s t r a t e d by examples.
Parameter v a r i a t i o n a n a l y s i s and r e l i a b i l i t y p r e d i c t i o n are t r e a t e d i n more d e t a i l
thanotherssincethese areas of a p p l i c a t i o n are p a r t i c u l a r l y s u i t e d t o computer
methods.Sec. 9 b r i e f l y summarizes t h es t a t e - o f - t h e - a r ti na u t o m a t i cd i g i t a l com-
putationemphasizingthoserecentdevelopmentsincommunicatingwiththecomputer
which make i t more s u i t a b l e t o e n g i n e e r i n g a p p l i c a t i o n .
Thecomputer o u t p u tc a nb en ob e t t e rt h a nt h e modelused t oo b t a i n it. Before
a computerprogramcanbewrittentoanalyze a piece of equipment, a c o n c e p t u a l model
o ft h a tp i e c eo fe q u i p m e n tm u s tb ef o r m u l a t e d .B e f o r ee x i s t i n gc o m p u t e rp r o g r a m sc a n

I
be used intelligently, the models they assumeandtherelationshipsofthosemodels
tothe equipmentwhich i s t ob ea n a l y z e dm u s tb e known. O f p a r t i c u l a ri m p o r t a n c e is
t h e knowledge of the parameter ranges over which the models assumed by a computer
program are v a l i d and how t h e s e r a n g e s r e l a t e t o a v a l i d model f o r t h e equipment
t ob ea n a l y z e d . A good d i s c u s s i o n o n t h e p r a c t i c a l a s p e c t s of modeling is p r e s e n t e d
i n Sec.2.1,Vol. I - P a r a m e t e rV a r i a t i o nA n a l y s i s of t h i s r e p o r t series.

2
2. FundamentalsofDigitalComputation
The purpose of t h i s s e c t i o n of t h e r e p o r t i s t o treat i n a s b r i e f a manner as is
consistent with clarity the fundamentals of the digital computerand its use.
2 . 1D i g i t a l Computer Concepts
A d i g i t a l computer system is.comprised of two elementswhichhave come t o b e
c a l l e dh a r d w a r ea n ds o f t w a r e .T h eh a r d w a r ec o n s i s t s of t h e p h y s i c a l p i e c e s of equipment,
viz,thecentralprocessor,thecard and t a p e r e a d e r s , t h e i n f o r m a t i o n s t o r a g e m e d i a ,
and t h e p r i n t e r s and p l o t t e r s . The s o f t w a r e c o n s i s t s o f a l l t h e computerprograms
which are a v a i l a b l e t o c a u s e t h e v a r i o u s p i e c e s ofequipment t o do u s e f u l t h i n g s .
A simplifiedblockdiagramof a stored-programelect.ronicdigitalcomputer is
shown i n F i g . 2-1. The o r g a n i z a t i o n a l s t r u c t u r e shown i n t h e f i g u r e i s common t o
e v e r y modern d i g i t a l computer although some computers may have more than one memory
u n i t ,a r i t h m e t i cu n i t , etc. Although d i g i t a l computersotherthanstored-program
electronic digital computers a r e of h i s t o r i c a l i n t e r e s t t h e y are n o t o f i n t e r e s t
i n modern e n g i n e e r i n g . I n t h i s r e p o r t when w e u s e t h e word computer w e s h a l l mean
stored-programelectronicdigitalcomputer.
The f u n c t i o n of a computer is t o t a k e d a t a v i a t h e i n p u t u n i t from t h e e x t e r n a l
w o r l d ,p e r f o r mc a l c u l a t i o n so n i t as s p e c i f i e d by t h e program stored in the memory
u n i t , and s u p p l y t h e r e s u l t s v i a t h e o u t p u t u n i t t o t h e e x t e r n a l w o r l d . I n a typical
installation the input unit is a punched c a r d r e a d e r w h i c h r e a d s t h e i n f o r m a t i o n on
t h ec a r d si n t ot h e memory u n i tu n d e rc o n t r o lo ft h ec o n t r o lu n i t . The t y p i c a l o u t p u t
unit is thelineprinter, whichproduces a p r i n t e d copy of t h e r e s u l t s of t h e c a l c u l a t i o n s .

n Control

F i g u r e 2-1. Basic Computer O r g a n i z a t i o n

3
 Computers are w i d e l y u s e d b o t h i n real-time o p e r a t i o n a n d i n o f f - l i n e o p e r a t i o n .
Although the terms real-time a n d o f f - l i n e are relative t o t h e a p p l i c a t i o n , t h e meaning-
ful distinction usually is t h a t i n t h e real-time a p p l i c a t i o n , t h e i n p u t d a t a mustbe
processedrapidlyand an output produced s o t h a t some k i n d o f r e s p o n s e c a n b e q u i c k l y
initiated. An example of t h e real-time a p p l i c a t i o n o f t h e d i g i t a l c o m p u t e r is i n
c o n j u n c t i o nw i t h a r a d a ri n s t a l l a t i o n .T h e r et h ei n p u td a t a comes f r o mt h er a d a r and
mustbeprocessedsufficientlyrapidlyto compute, f o r example,guidance commands f o r
a missile l a u n c h e d t o i n t e r c e p t a n a t t a c k i n g a i r c r a f t . We will n o t d i s c u s s i n t h i s
reporttheuse of d i g i t a l computers i n such real-time applications.
ReferringagaintoFig. 2-1 w e c o n s i d e r b r i e f l y t h e f u n c t i o n of e a c h of t h e b l o c k s
shown. First,the memory u n i t s e r v e s as s t o r a g e f o r ( 1 ) t h e programwhich is t ob e
e x e c u t e d , (2) t h e i n p u t d a t a u n t i l i t i s needed f o rp r o c e s s i n g , (3) i n t e r m e d i a t e r e s u l t s
d u r i n gt h ee x e c u t i o no ft h ep r o g r a m , and ( 4 ) t h e f i n a l r e s u l t s u n t i l t h e y are ready
f o ro u t p u t . The memory u n i t t y p i c a l l y is a p r i n c i p a le l e m e n to ft h ec o m p u t e r ;t h e
costandspeedofthemodemdigitalcomputer are l a r g e l y g o v e r n e d by t h e c o s t and
speedofthe memory. It i s n o t uncommon f o r t h e c o s t of t h e memory t o approach
the cost of a l l t h e o t h e r u n i t s combined.
The memory c o n t e n t s are s t o r e d i n t h e formofbinarydigits(bits) which are
grouped i n t o b l o c k s of s u f f i c i e n t s i z e f o r t h e number r a n g e a n d p r e c i s i o n r e q u i r e m e n t s
f o r which thecomputer is designed.Such a b l o c k of b i n a r y d i g i t s i s c a l l e d a memory
word. I n computers i n common u s e t o d a y t h e memory word v a r i e s from12 b i t s up t o 60
b i t s , whichcorrespondsto a decimal number r a n g e of 4000 t o 1 0
18
. The number of words
t h a t a computer memory may s t o r e a l s o v a r i e s w i d e l y andrangesfrom1000words up t o
6
1 0o r morewords.
A s s o c i a t e dw i t ht h e memory u n i t are two r e g i s t e r s
*. These are t h e memory a d d r e s s
register a n d t h e memory data register. When i t is d e s i r e d t o s t o r e a number i n memory
o r r e t r i e v e i t from memory, i t is n e c e s s a r y t o g i v e t h e l o c a t i o n of t h e p a r t i c u l a r
memory word d e s i r e d . The memory a d d r e s s r e g i s t e r i s used t o d e s i g n a t e t h e a d d r e s s ,
i.e. t h e l o c a t i o n , o f t h e word i n memory. When t h e command i s g i v e n by t h e c o n t r o l
unit to store or retrieve a word from memory, t h e memory a d d r e s s r e g i s t e r is usedto
d e s i g n a t et h ea d d r e s s .T h e r e are as many u n i q u ea d d r e s s e s , i. e. l o c a t i o n s a t which
a number c a n b e s t o r e d , i n t h e memory as t h e number of wordswhich t h e memory is
c a p a b l eo fs t o r i n g .
The memory d a t a r e g i s t e r is used as a n i n t e r m e d i a t e s t o r a g e when a word i s going
from t h e a r i t h m e t i c u n i t o r t h e i n p u t u n i t t o t h e memory. To s t o r e a word i n memory,

*A r e g i s t e r i s a t e m p o r a r ys t o r a g ed e v i c e . I t t y p i c a l l yc a ns t o r eo n e memory
word.

4
I

the word is placed into the memory data register, and the address at which it is to
be stored is placed into the memory address register. Then the store command generated
by the control unit causes the word
be to
stored at the specified address. When a
data word is to be retrieved from memory, the address ofisthe
again
word
placed into
the memory address register, and the fetch command from the control unit causes the
word to be transferred from the specified address memory
in theto the memory data
register.
The arithmetic unit performs an arithmetic (or logic) operation as specified by
the program between a word contained
in a register in the arithmetic unit called the
accumulator and a word fetched from memory into the memory data register. This des-
cription holds for the single address computer. The term single address means simply
that a single program step (which also is stored as a word in memory but is called
an instruction e)
specifies the address of only one data word in memory.
The second
word to be used
in an operation is contained in the accumulator register in the arith-
metic unit. Althought some computers specify more than one address in one instruction
word, the single address computer organization is the most widely used.
In the single address computer, the accumulator register contains one operand
*
for an operation, with the other operand being first in memory and later in the
memory data register. The resultof an operation usually ends up in the accumulator.
Data words can be fetched from memory to the accumulator or stored from the accumulato
into the memory. Except when the computer instruction specifically calls for it, the
'contents of the accumulator are not disturbed by an operation.
The control is the logic complex which determines which operation is to be
performed at what time and what sequence of elementary logic steps accomplishes the
operation. The control unit contains two very important registers--the program register
(also called the instruction counter) and the instruction register.
The program stored in the computer memory unit consists of a sequence of instructions
which the computer is to perform. The program is stored in the memory in the proper
sequence: the first instruction is stored in some location
n, the second stored in
location n+l, etc. The function of the program register is to keep track of the loca-
tion from which the next instruction is to be fetched; it does this by counting the
instructions as they are performed. Unless specifically requested toso do
by a specific
instruction, the program will proceed in sequence
by picking up its instructions from
successive memory addresses.

*An operand is any single-word quantity which

is operated upon by the computer.

5
 The function of the instruction register is t o t e m p o r a r i l y s t o r e e a c h i n s t r u c t i o n
to enable the control unit to decode i t and i n i t i a t e a n d p r o p e r l y time t h e s e q u e n c e
o fe l e m e n t a r yl o g i cs t e p sw h i c hi m p l e m e n t st h ei n s t r u c t i o n . I t is a f u n d a m e n t a lf a c t
thatthe memory c o n t a i n s b o t h t h e i n s t r u c t i o n s t o b e e x e c u t e d , (i.e. t h e program)and
the data on which t h e i n s t r u c t i o n s are t o o p e r a t e .
The two k i n d s o f s t o r e d w o r d s ( i n s t r u c t i o n s a n d d a t a ) are t r e a t e d i n two e n t i r e l y
d i f f e r e n t ways. An i n s t r u c t i o n is t r a n s f e r r e d t o t h e i n s t r u c t i o n r e g i s t e r where i t
i s examined by t h e c o n t r o l u n i t t o d e t e r m i n e :
( 1 )w h a to p e r a t i o n( a d d ,s u b t r a c t ,l o g i c ,e t c . ) is required,
(2) where thesecondoperand is located, i.e., t h ea d d r e s s of thesecond
operand , and
(3) where t h e r e s u l t o ft h eo p e r a t i o ns h o u l db ep l a c e d .
I f , as i s u s u a l l y t h e case, one of t h eo p e r a n d s i s c o n t a i n e di nt h e memory, t h e n t h i s
operandaddress is containedintheinstruction word l o c a t e d i n t h e i n s t r u c t i o n r e g i s t e r .
T h i sa d d r e s s is f u r n i s h e d t o t h e memory a d d r e s s r e g i s t e r at thecorrect time as s p e c i f i e d
by t h e c o n t r o l u n i t . The r e s u l t of t h eo p e r a t i o nu s u a l l yg o e si n t ot h ea c c u m u l a t o r .
The i n p u t u n i t and o u t p u t u n i t h a v e a s s o c i a t e d w i t h them a d a t a r e g i s t e r and a n
addressregisteranalogoustothe memory a d d r e s s r e g i s t e r and memory d a t a r e g i s t e r
of t h e memory u n i t . Data comingfroman e x t e r n a ld e v i c e i s p l a c e di n t ot h ei n p u t -
output(I/O)dataregister and later t r a n s f e r r e d i n t o t h e a c c u m u l a t o r f o r u s e i n s i d e
thecomputer. Data g o i n gt oa ne x t e r n a ld e v i c e i s t r a n s f e r r e d from theaccumulator
tothe 1/0 d a t a r e g i s t e r fromwhich i t is removed by t h e 1/0 d e v i c e . S i n c e t y p i c a l l y
s e v e r a li n p u t - o u t p u td e v i c e s are c o n n e c t e dt ot h ec o m p u t e r ,i n p u t - o u t p u ta d d r e s s e s
m u s tb es p e c i f i e dt oi d e n t i f yw h i c h 110 d e v i c e is r e q u e s t e d . The f u n c t i o n of t h e
I/o a d d r e s s repister i s t od e s i g n a t et h ea d d r e s so ft h e I/O d e v i c e ; t h e a d d r e s s of
the If0 device i s n o t h i n g more t h a n a number which i t has been given to uniquely
i d e n t i f y it.
A c o m p u t e rc a np e r f o r mo n l yt h eo p e r a t i o n sw h i c hh a v eb e e nb u i l ti n t o it. The
l i s t of o p e r a t i o n s w h i c h a computercanperform is c a l l e d t h e i n s t r u c t i o n r e p e r t o i r g
ofthecomputer. Any programwhichcanbeexecuted by a computer i s made up ofonly
thoseinstructionscontainedinthatcomputer'sinstructionrepertoid.
An i n s t r u c t i o n is a s t e p i n a programbut we wish to indicate in detail what
c o m p r i s e sa ni n s t r u c t i o n .F o rp u r p o s e so fd i s c u s s i o nt h ef o l l o w i n gd e s c r i p t i o n of
a ni n s t r u c t i o n is r e f e r e n c e d t o a s i n g l ea d d r e s sc o m p u t e r . A c o m p u t e ri n s t r u c t i o n is
made up of t h r e e b a s i c p a r t s :
The o p e r a t i o n code (opcode) is t h a t p a r t ofan i n s t r u c t i o n which s p e c i f i e s t o
t h ec o n t r o lu n i tw h i c ho p e r a t i o n i s t ob ep e r f o r m e d( a d d ,s u b t r a c t ,t r a n s f e rd a t a
t o o r from memory, etc.)

6
 I
I "

The i n s t r u c t i o n m o d i f i e r is a group of b i t s which f u r t h e r s p e c i f i e s how t h e i n s t r u c -

t i o n is t o b e p e r f o r m e d . F o r e x a m p l e , t h e a d d o p e r a t i o n o r d i n a r i l y r e s u l t s i n t h e
Sum b e i n gp l a c e di nt h ea c c u m u l a t o ro n l y . A modifiertothe addopcodemightspecify
that the result of the add operation also be placed into a memory l o c a t i o n .
The addressofthesecondoperand is the third basic part ofaninstruction. The
a d d r e s s i s simply the number of t h e memory l o c a t i o n w h i c h c o n t a i n s t h e d a t a t o b e
o p e r a t e d on as s p e c i f i e d b y t h e opcode.
Thecomputerhas two c h a r a c t e r i s t i c s which make i t an e x c e e d i n g l y p o w e r f u l a i d
t o p r o b l e ms o l v i n g .F i r s t ,t h ec o m p u t e rc a np e r f o r mo p e r a t i o n s( a l b e i ts i m p l e )
exceedinglyrapidly. I t i s n o t uncommon f o r a l a r g e - s c a l e moderncomputer to be able
t op e r f o r m ,f o re x a m p l e ,o n em i l l i o na d d i t i o no p e r a t i o n si no n es e c o n d
*. Fundamental
to the ability to dosimpleoperationsexceedinglyfast is t h e a b i l i t y t o o b t a i n t h e
d a t a a t a na d e q u a t e l yr a p i d rate. The a v a i l a b i l i t y of t h e d a t a a t such a rate i m p l i e s
that both the instructions for operating on t h e d a t a and t h e d a t a i t s e l f mustbe
stored in the computer memory.
The s e c o n dc h a r a c t e r i s t i co ff u n d a m e n t a li m p o r t a n c ei nt h ec o m p u t e r is its
a b i l i t y t o performthe same s e q u e n c e o f o p e r a t i o n s a n a r b i t r a r y number of t i m e s , ex-
ceptthatthesequence i s performedeach time on a d i f f e r e n t s e t of d a t a - - t h i s is the
a b i l i t y of thecomputertomodify i t s own program. I t m i g h ta p p e a rt h a tt oi n s t r u c t
a computer t o p e r f o r m t h e o p e r a t i o n s n e c e s s a r y t o add by p a i r s two t a b l e s o f 1 0 0
numbers each would r e q u i r e 200 o r more i n s t r u c t i o n s . On t h e c o n t r a r y , i t i s a simple
matter t o p u t i n s t r u c t i o n s i n t h e programwhichmodify the instruction addresses in
a way t o s t e p t h r o u g h t h e p a i r s o f numbers i n t h e t a b l e s and make t h e t o t a l number
o fr e q u i r e di n s t r u c t i o n ss o m e t h i n gl i k et e n .
2.2 Computer
Programming
Languages
Inthefinalanalysis a d i g i t a l c o m p u t e rc a no n l yr e c o g n i z eb i n a r yp a t t e r n s . Thus
t h e r e are severalprogramsbetweenthecomputer programmer u s i n g FORTRAN ( o r a n o t h e r
h i g h - l e v e l programming l a n g u a g e ) a n d t h e a c t u a l e x e c u t i o n by thecomputerofthe
o p e r a t i o n sr e q u e s t e d by t h e programmer i n h i s FORTRAN program. T h r e el e v e l so f
computerlanguages are i n wideusetoday:assemblylanguapes,procedure-oriented
languapessuch as FORTRAN, andproblem-orientedlanguagessuch as t h e i n p u t l a n g u a g e
f o r a u t o m a t i cc i r c u i ta n a l y s i sp r o g r a m s .

*Theresponse t i m e of t h e l o g i c d e v i c e s i n t e r n a l t o a modern computer is a few

nanoseconds,which is a n i n t e r e s t i n g c o n t r a s t t o t h e f e wm i l l i s e c o n d sr e s p o n s e time
of t h e n e u r o n s o f t h e c o m p u t e r u s e r .

7
 The assemblylanguage
* is the first level ofcomputerlanguage removed f r o mt h e
b i n a r yp a t t e r n sw h i c ht h ec o m p u t e rd i r e c t l yr e c o g n i z e s .C o n s i d e rt h e add o p e r a t i o n .
The binary pattern for the add operation(which i s t h e add op codelfor a particular
computermightbe1000.Beforethecomputercanactuallyexecutean add o p e r a t i o n ,
i t musthave i n t h e op c o d e p o r t i o n o f its i n s t r u c t i o n r e g i s t e r t h e b i n a r y p a t t e r n
1000. It a l s o musthave, intheaddressfield o ft h ei n s t r u c t i o n word, t h e b i n a r y
p a t t e r n whichgivesthelocation of t h e memory word c o n t a i n i n g t h e d a t a w h i c h is t o
b e added t ot h ec o n t e n t so ft h ea c c u m u l a t o r . The assemblylanguageenablesthepro-
grammer t o u s e a s u g g e s t i v es e q u e n c e of l e t t e r s c a l l e d a n i n s t r u c t i o n mnemonic, f o r
example ADD i n t h e case of t h e a d d o p e r a t i o n , t o s p e c i f y t h a t a n a d d i t i o n is t o b e
performed.Before t h i sa d d i t i o no p e r a t i o nc a l l e df o r by theassemblylanguagepro-
graqcanbeperformedinthecomputer, i t mustbeprocessed by anothercomputer program--
calledtheassembler--whichhastheabilitytointerpretthe l e t t e r s ADD as t h e op code
1000 f o r t h e add o p e r a t i o n . I f w e wish t o add t h e numbers X and Y , t h e a v a i l a b i l i t y
of t h e assemblylanguageenablesusto write a sequence of i n s f r u c t i o n s w h i c h l o a d s
t h ea c c u m u l a t o rw i t h X, adds Y , and s t o r e s t h e r e s u l t a t a d e s i r e dl o c a t i o n 2. Such
a sequence i s
LDA X
ADD Y
ST0 2
where LDA, ADD, and ST0 a r e r e s p e c t i v e l y t h e mnemonics f o rl o a d i n gt h ea c c u m u l a t o r
f r o mt h e memory, a d d i n gt ot h ea c c u m u l a t o r ,a n ds t o r i n gt h ec o n t e n t so ft h ea c c u m u l a t o r
inthe memory. Eachof t h e l e t t e r s X, Y , and Z r e p r e s e n t st h es y m b o l i ca d d r e s so f a
memory word. The a s s e m b l e ri na d d i t i o nt oc o n v e r t i n gt h ei n s t r u c t i o n mnemonics t o t h e i r
binaryequivalents,allocates memoryh:vds and convertseachsymbolicaddressused
i n anassemblylanguageprogram t o a f i x e db i n a r y memory a d d r e s s . Thus assembly
language programming c o n t r a s t s t o h a v i n g t o write t h e b i n a r y p a t t e r n s f o r e a c h computer
i n s t r u c t i o n and t o a l l o c a t e memory l o c a t i o n s by w r i t i n g a b i n a r y memory a d d r e s s f o r
e a c hd a t a word used i n t h e program.
The procedure-orientedlanguape, of which FORTRAN
**i s thebest known andmost
widelyusedexample,effectivelyremovesthe programmer o n e l e v e l f u r t h e r from t h e
t e d i o u st a s ko f programming t h e c o m p u t e rw i t hb i n a r yp a t t e r n s .T h u s ,w h e r e a st h r e e
a s s e m b l yl a n g u a g ei n s t r u c t i o n s were r e q u i r e d t o s p e c i f y t h e a d d i t i o n o f X andYand

*Assembly
language i s a l s o c a l l e d m a c h i n e l a n g u a g e , s i n c e t h e d e t a i l s o f a n a s s e m b l y
language are h i g h l y d e p e n d e n t o n t h e d e t a i l s of t h e s p e c i f i c m a c h i n e ( t h e s p e c i f i c c o m p u t e r )
onwhich i t i s u s e d . O r i g i n a l l y , m a c h i n e l a n g u a g e m e a n t t h e b i n a r y p a t t e r n s d i r e c t l y
recognized by a computer.
**FORTRAN i s a contractionof"formulatranslation".
store the result in the memory a t l o c a t i o n Z, t h e FORTRAN s t a t e m e n t f o r a c c o m p l i s h i n g
t h i s wouldbesimply 2 = X + Y. Theprogramwhichprocessesthe FORTRAN s t a t e m e n t
( c a l l e d t h e FORTRAN compiler) would produce the same s e q u e n c e o f b i n a r y p a t t e r n s t h a t
t h ea s s e m b l yl a n g u a g ei n s t r u c t i o n sp r o d u c e . Whereas i n w r i t i n g i n assemblylanguage
one statement must be written for each instruction to be executed, a FORTRAN s t a t e m e n t
(and i n g e n e r a l a n y p r o c e d u r e - o r i e n t e d l a n g u a g e s t a t e m e n t ) w i l l produceseveralcomputer
instructions,typicallyfourorfive.
An advantage of procedure-oriented languages which is probably more important
than their ease of use by t h e programmer is t h a t a procedure-orientedlanguageprogram
is nearlymachineindependent,indramaticcontrasttotheprogramwritteninassembly
language. Thus a programwhich is w r i t t e n i n FORTRAN c a n b e i n t e r p r e t e d , v i a t h e
FORTRAN compilerofanycomputerwhichhasone,andthenexecuted on t h a t c o m p u t e r ,
withonlyminorprogramchangesbetweendifferentcomputers. A s p e c i f i c computer
a l m o s tn e v e rs t a y si n a particularinstallationfor more t h a n a fewyears. The u s e
of procedure-orientedlanguage programming is t h e o n l y e f f e c t i v e way t o p r e v e n t l o s i n g
thelargeinvestmentin programming time andchecked-outprograms f o rt h eo l dc o m p u t e r
when t h e new computer is i n s t a l l e d .
The problem-oriented language i s thenewestandin many ways t h e mostpowerful
computerlanguage. A s i n g l es t a t e m e n ti n a p r o b l e m - o r i e n t e dl a n g u a g em i g h tr e s u l t
i nt h ee x e c u t i o no f up t os e v e r a lt h o u s a n dc o m p u t e ri n s t r u c t i o n s .P r o b l e m - o r i e n t e d
languages are d i s c u s s e d i n l a t e r s e c t i o n so ft h er e p o r t .I ne s s e n c et h e yc o n s i s to f
theinputlanguagestospecialprogramswrittentoaidinspecificproblem areas, e.g. ,
problems i n n e t w o r k a n a l y s i s .
In the final analysis, the computercan do no more and no less t h a n p r e c i s e l y
what i t i s i n s t r u c t e d t o do v i a t h e program.Givenadequatelycleverpeoplepreparing
andusingthecomputerprograms,thecomputercanindeed do some v e r y i m p r e s s i v e t h i n g s .
As an.aid to design for reliability, the computerenablesequipmentdesignersto con-
d u c t many moreand more t h o r o u g h a n a l y s e s o f t h e i r d e s i g n s t h a n would b e p o s s i b l e by
anycombinationof hand c a l c u l a t i o n and l a b o r a t o r ye x p e r i m e n t a t i o n . However, i t is
up t o t h e computeruserstoexaminetheoutputfromtheprogramsthey are u s i n g , t o
interpret the computer r e s u l t s , andthemselves t o make t h e c o r r e c t i o n s anddesign
m o d i f i c a t i o n sw h i c ht h e yd i s c o v e rv i ac o m p u t e ra n a l y s i s . Thecomputer d o e sn o t by
any s t r e t c h of t h e i m a g i n a t i o n remove t h e n e e d f o r good e n g i n e e r i n g and c l e a r t h i n k i n g
i n t h e development and design of reliable equipment.
2.3 Basic MathematicalProblemsthat Can B e Solved by a Computer
Problem s o l v i n g is a n e s s e n t i a l p a r t o fe n g i n e e r i n gd e s i g n . Some of t h e problems
are v e r y s i m p l e from a c o m p u t a t i o n a l s t a n d p o i n t , r e q u i r i n g o n l y a slide rule, a
p e n c i l , and a p i e c e o f p a p e r , w h i l e o t h e r p r o b l e m s r e q u i r e a team of engineers working

9
many d a y s o r p e r h a p s y e a r s . The l a t t e r problems were a t t a c k e d by approximations
b a s e do ns i m p l i f y i n ga s s u m p t i o n s when d i g i t a l computers were n o t a v a i l a b l e . However,
i t is now p r a c t i c a l t o e v a l u a t e t h e a d e q u a c y o f s u c h a s s u m p t i o n s a n d d e l v e i n t o s y s t e m
analysis problems which would h a v e b e e n i m p r a c t i c a l o n l y a few y e a r s ago.
Solving a particular engineering problem on a c o m p u t e r u s u a l l y r e q u i r e s t h e u s e
o fs e v e r a lb a s i cm a t h e m a t i c a lt e c h n i q u e s .F o re x a m p l e ,s u p p o s et h a t we w i s h t o o b t a i n
t h e minimum v a l u e o f a p a r t i c u l a r known f u n c t i o n f ( x ) on a c e r t a i n i n t e r v a l [ a , b ] .
I n some cases t h e d e r i v a t i v e f u n c t i o n c a n b e w r i t t e n w i t h o u t d i f f i c u l t y , t h e r e s u l t i n g
equationsolvedforthezeros,andthe.solutionstestedtodeterminewhichvalueof
the independent variable yields the minimum v a l u e o f t h e r e s p o n s e o r p e r f o r m a n c e
variable. However, i n some p r o b l e m st h ew r i t i n go ft h ed e r i v a t i v et a k e sc o n s i d e r a b l e
time and i t s e v a l u a t i o n a g r e a t d e a l l o n g e r time t h a n t h e e v a l u a t i o n o f t h e o r i g i n a l
f u n c t i o n , and o f t e n t h e e q u a t i o n o b t a i n e d by e q u a t i n g t h e d e r i v a t i v e t o z e r o is hard
t os o l v e . Hence a computer is u s e d t o a i d i n t h e a n a l y s i s .
Again t h e r e are many avenuesofattack on t h e problem. One approach i s t o
evaluatethefunctionf(x) a t a singlevalueof x withinthegiveninterval and
t h e n select a n o t h e r x v a l u e a t some p r e d e t e r m i n e d d i s t a n c e f r o m t h e f i r s t p o i n t and
compare t h e two v a l u e s . I f t h e v a l u e of t h e f u n c t i o n a t t h e s e c o n d p o i n t i s less t h a n
at the first point, take i t as a new r e f e r e n c e p o i n t andproceed t o a thirdpoint, etc.
Insuch a process the interval of step size between successive x's mustbedecreased
i n a s y s t e m a t i c manner when noimprovement results from increasing or decreasing x
by t h e p r e s c r i b e d step size. Sucha procedure w i l l u l t i m a t e l y l e a d t o anadequate
solutionof a problemof a localminimq,andinthe case of a convexfunctiononthe
*
interval an absolute minimumyas s e e n i n t h e f i g u r e below.

V Local
Minimum

Absolute
Minimum

F i g u r e 2-2. Minima of a F u n c t i o nf ( x )
*L i n e a r i n t e r p o l a t i o n n e v e r u n d e r e s t i m a t e s t h e
real v a l u e ofaconvex function at
theinterpolatedpoint.For a mathematicaldefinitionof a convex f u n c t i o n , see Ref. 2.2.

10
I

Anotherattack on t h e p r o b l e m is t o select t h r e e p o i n t s o n t h e i n t e r v a l [ a , b ] ,
fitthecorrespondingy's byparabola,and estimate t h e l o c a t i o n o f t h e v e r t e x . Then,
select t h r e e new p o i n t s i n t h e n e i g h b o r h o o d o f t h i s vertex and r e p e a t t h e a b o v e ;
eventually the location of t h e l o c a l minimum p o i n t is d e t e r m i n e d t o w i t h i n t h e d e s i r e d
d e g r e e of p r e c i s i o n . T h i s a p p r o a c h r e q u i r e s t h e e v a l u a t i o n of t h e f u n c t i o n a t t h r e e
p o i n t s and t h e s o l u t i o n o f a set of three linear simultaneous equations for each
iteration. It a l s o r e q u i r e s t h e p r o v i s i o n of a l o g i c a l p r o c e d u r e f o r a l t e r i n g t h e
s t e p s i z e as t h e i t e r a t i o n s c o n v e r g e t o w a r d t h e s o l u t i o n .
If the function is convexandonlyoneindependentvariable is i n v o l v e d , t h e r e
is a n e a r optimum p r o c e d u r e f o r f i n d i n g t h e minimum u s i n g t h e p r o p e r t i e s of t h e
F i b o n a c c i numbers 1, 1, 2, 3, 5 , 8, 13, 21, ..., whereeach number i nt h es e q u e n c e
is o b t a i n e d by a d d i n g t h e two p r e v i o u s n u m b e r s , t h a t is

Thisprocedure is c o n s i d e r e d i n Ref. 2-2 u n d e rt h eb a s i cp r o b l e m so fo p t i m i z a t i o n .

Also see Ref. 2-3 f o r a m a t h e m a t i c a l t r e a t m e n t o f t h i s s u b j e c t .
Inthefollowingsectionsarepresented some of t h e b a s i c p r o b l e m t y p e s , some
of t h e a p p r o a c h e s t o s o l u t i o n , and t h e r e l a t i o n of t h e s e b a s i c p r o b l e m s t o t y p i c a l
engineeringproblemsviaparticularcomputerprograms.Thisapproach was s e l e c t e d
toavoid some of theredundancywhich would occur as a r e s u l t of t r e a t i n g p r o b l e m s
inelectronics,orpropulsion,orstructuresasseparate problems when i n f a c t t h e y
may be a l l of t h e same b a s i c problem a r e a .
Function Evaluation
The f i r s t problemtype i s one of evaluating a functionofoneorseveralvariables
d e f i n e d by

y = f ( x1' x29 ". 9

= f(xJ

where x= (xl, ..., xn ) and x

i
is t h ei - t hv a r i a b l e .F o rs i m p l ef u n c t i o n s a computer
is n o t n e e d e d t o s o l v e f o r y forgivenvalues of t h e xi;however, if the operation
is t o be repeated frequently or if the function is complex t h e u s e o f a computer i s
.oftenjustified. The d i s p l a y o f t h e o u t p u t i n a t a b l eo rg r a p hf o r m i s importantfrom
t h eu s e rs t a n d p o i n t .I ft h ef u n c t i o n is animportantone a t a b l e of v a l u e s f o r f u t u r e
u s ec a nb ep r e p a r e df o rd i f f e r e n tv a l u e s of 11. It i s o b v i o u s t h a t a computercanbe
used to obtain reams o f p a p e r c o n t a i n i n g n u m e r i c a l v a l u e s o f y for various combinations
of x
i'
i=1, ..., n. However, t h eo b j e c t i v eo ft h e problemand t h eu s e st ob e made
o ft h er e s u l t ss h o u l db et h o r o u g h l yc o n s i d e r e dp r i o rt oc o m p u t a t i o n . There is no

11
need t o t a b u l a t e a function which can be computed a l m o s t as r e a d i l y byhand as one
can l o c a t e t h e t a b l e a n dt h e nl o o k i t up.Although thisstatement seems obvious i t
is p o s s i b l et ol o c a t ee x a m p l e so fs u c hf u n c t i o n st a b u l a t e di nt h el i t e r a t u r e .A l s o ,
t h es e l e c t i o n o ft h ev a l u e so ft h e x
i
i=l, ..., n a t which t o compute t h e y s i s
animportantaspectoftheproblem.
I ne n g i n e e r i n ga p p l i c a t i o n st h ep e r f o r m a n c eo r some f i g u r e - o f - m e r i t (FOM) ofan
equipmentcanoftenbeexpressed as a f u n c t i o n of t h e c h a r a c t e r i s t i c s of i t s p a r t s
and t h ei n p u t s ,e n v i r o n m e n t s ,l o a d s , etc. Thus t h e FOM may b eo b t a i n e df o rv a r i o u s
v a l u e so ft h ev a r i a b l e s which i n f l u e n c e i t . Computationof s t a t i c anddynamicresponses
withcircuit and s t r u c t u r a l e q u a t i o n s are t y p i c a l examples i n e n g i n e e r i n g .
FunctionalEquation
N e x t considertheinverseproblemofsolvingfor 5 given y, i. e . , i f

d e t e r m i n e 5 s u c ht h a tf ( 5 ) = y o ,w h e r et h es o l u t i o n ( s ) w i l l bedenoted by x
a
For .
example, w e may h a v e a n a l g e b r a i c e q u a t i o n i n o n e v a r i a b l e x and wish to solve for
thevaluesof x a t which t h e c u r v e c o r r e s p o n d i n g t o t h e e q u a t i o n y = f(x) crosses
orintersectsthe x axis ( l i n e y E 0 ) . We may wish t o o b t a i n t h e e x t r e m e p o i n t s
(maxima, minima, p o i n t s o f z e r o d e r i v a t i v e ) f o r f ( x ) when t h e d e r i v a t i v e f u n c t i o n
f ( x )c a nb er e a d i l yo b t a i n e d .I ng e n e r a lt h ep r o b l e m may r e q u i r e t h e u s e ofan
i t e r a t i o nt e c h n i q u e ,s u c h as t h e Newton-Raphson method
* o fs o l v i n ga ne q u a t i o n by
usingtheconstruction of s u c c e s s i v e t a n g e n t s t o t h e c u r v e a t pointsapproachingthe
solution.
A typicalengineering exampleoftheaboveproblem is tofindtheparametervalues
y i e l d i n g a g i v e nl e v e lo fp e r f o r m a n c e . It i s p o s s i b l et oo b t a i nc o n t o u r s of e q u a l
p e r f o r m a n c ev a l u e so ft h e set of a l l v a l u e s o f t h e i n d e p e n d e n t v a r i a b l e s c o r r e s p o n d i n g
t o y = yo , yl, ..., ym.Such a s e t ofcontours i s i n d i c a t e di nF i g . 2-3. Such
techniquescanbehelpfulindeterminingtheoperatingconditionsyieldingthedesired
performance. The abovetechnique becomes v e r y h e l p f u l when two o r more dependent
o rp e r f o r m a n c ev a r i a b l e s are beingconsidered.Forexample, i nF i g . 2-4 two v a r i a b l e s
are shown and t h er e g i o no fo p e r a t i o nd e f i n e d by t h e set o ft h e x i=l,2 ,f o r which
i
y1 1 30, y2 5 20. The s h a d e dr e g i o np r o v i d e s a r e g i o no fo p e r a t i o nw h i c hs a t i s f i e s
t h eg i v e nc o n s t r a i n t s .F u r t h e rd i s c u s s i o no fs u c ha na p p r o a c ha n dp r a c t i c a lp r o b l e m s
associatedwith i t are i n Vol. I - Parameter V a r i a t i o n A n a l y s i s o f t h i s series.

*There are numerous t e x t s on standardnumericalmethods.Refs.2-4 and 2-5 are

good s t a r t i n g p o i n t s .

12
 F i g u r e 2-3. TypicalPerformanceContours

-"2
A

yl=30

y =20
2

I
* x1

F i g u r e 2-4. RegionofDesiredPerformance

13
 I f a system of equations is i n v o l v e d t h e p r o b l e m may have a s i n g l e s o l u t i o n o r
a 'multiplesolutiondepending on t h e d e g r e e o f t h e e q u a t i o n s , t h e numberof equations
relativetothe number o f unknowns, etc. Many problems i n real w o r l d a p p l i c a t i o n s
r e s u l t i n a systemofequationstobesolvedforthevalueorvaluesofthe unknown
v a r i a b l e s which s a t i s f y s p e c i f i e d c o n d i t i o n s . Some of t h e s ep r o b l e m s will b ec o n s i d e r e d
later.
Functional Approximation
Anotherimportantproblem i n computer a p p l i c a t i o n is t h e u s e of f u n c t i o n a l
a p p r o x i m a t i o n st of u n c t i o n s whichcannot'beexpressed i n a c l o s e df o r m y e .g . , some
i n d e f i n i t ei n t e g r a l so rt h e sum ofan i n f i n i t e series. Forexample,theapproximations
X
t o s i n x and e canbe in the formof a T a y l o r series o ro r t h o g o n a l .p o l y n o m i a l ss u c h
a s Chebyshev,Legendre,andHermitianpolynomials.In many a p p l i c a t i o n s a f i n i t e
T a y l o r series approximation is t o beused. On t h eo t h e rh a n d ,e x t r e m e l ya c c u r a t e
a p p r o x i m a t i o n sa r es o m e t i m e sn e e d e d ,s u c ha sf o rt h ec u m u l a t i v ep r o b a b i l i t yi n t e g r a l
o ft h eG a u s s i a nd i s t r i b u t i o n .R a t i o n a li n t e g r a lf u n c t i o n sa r eo f t e n used i na p p r o x i m a t i n g
suchcurves.SeeRef.2-5forexamplesofapproximationsto a v a r i e t y o ff u n c t i o n s .
One u s e f u l a p p l i c a t i o n i n e n g i n e e r i n g p r o b l e m s is r e d u c i n g a complex f u n c t i o n t o
a linearor , when n e c e s s a r y ,t o a seconddegreeapproximation. Suchanapproach is
useful in deriving the properties of t h e d i s t r i b u t i o n o f a performancevariable y
i n terms of t h e c h a r a c t e r i s t i c s o f t h e d i s t r i b u t i o n s of t h e i n d e p e n d e n t v a r i a b l e s .
It i s a l s o a p p l i e d o f t e n i n c o n s t r u c t i n g c o n t o u r s a n dp e r f o r m i n gs e n s i t i v i t ya n a l y s e s .
A linearapproximation i s most o f t e n s u f f i c i e n t o v e r t h e r e g i o n of i n t e r e s t .
Thisproblemtypeleadslogicallyintotheproblemareaofcurvefitting which
i s discussedbelow. The two problems a r es e p a r a t e dh e r eb e c a u s et h ef i r s t problem
type deals with a known model d e f i n e d e x p l i c i t l y s u c h a s
X
y = e

or

, x

or only implicitly such as

f(t, y, dy/dt, ...) = 0.

The curve f i t t i n g problem on t h e o t h e r hand t r e a t s a g i v e n model form w i t h unknown

constantstobedeterminedfromgivendata.

14
I

Curve F i t t i n g
Suppose t h a t i n s t e a d o f b e i n g g i v e n a f u n c t i o n as suggested above one is g i v e n
a set of v a l u e s yi and t h e c o r r e s p o n d i n g xi or x i n t h e case of s e v e r a l i n d e p e n d e n t
"i
v a r i a b l e s .F o ro n ei n d e p e n d e n tv a r i a b l ea n do n ed e p e n d e n tv a r i a b l e , a c u r v e may b e
fitted to the data freehand. If however w e have some knowledgeconcerningtheunder-
l y i n g mechanism (a modelform)andwish t o estimate c e r t a i n c o n s t a n t s o r p a r a m e t e r s
of t h e model, a more a p p r o p r i a t e p r o c e d u r e would b e t o estimate t h e p a r a m e t e r s b y a
mathematicalproceduresuch as t h e method of l e a s t s q u a r e s . Even when t h e modelform
i s n o t known, t h e r e i s o f t e n c o n s i d e r a b l e a d v a n t a g e i n f i t t i n g t h e c u r v e by a mathe-
matical interpolation or a graduation formula such as a l i n e a r o r second degree
function in x orpossiblyinl/xdependingonthenatureofthegivendata. Such a
predictionequation is s a t i s f a c t o r y o n l y i n t h e r e g i o n of t h e g i v e n d a t a u n l e s s
t h e o r e t i c a l knowledge is a v a i l a b l e t o a l l o w c o r r e c t e x t r a p o l a t i o n beyond t h e r e g i o n
o fe x p e r i m e n t a lr e s u l t sg i v e n by t h e d a t a .
Anothercloselyrelatedtechniqueforfitting a curve i s s m o o t h i n g t h e d a t a .
Smoothing t h e d a t a i s based on t h e f i t t i n g o f p o l y n o m i a l s t o a set o f s u c c e s s i v e d a t a
p o i n t s and c a l c u l a t i n g t h e "smoothed" p o i n t s .F o re x a m p l e ,s u p p o s et h a t 2t +1
successive equally-spaced points, (t = 1,2, ...) a r e s e l e c t e d and a polynomial of
degreethreefittedtothesepoints. Then t h e smoothed v a l u e of y is g i v e n by

y2
*= -
1 ( - 3 ~ ~12y1
35
+ + 17y2 + 12y3 - 3y4)

where y
0'
yl, ..., y 4 a r e f i v e c o n s e c u t i v e v a l u e s of y.
The least s q u a r e s t e c h n i q u e h a s t h e most u s e f u l a p p l i c a t i o n when f i t t i n g a curve
t o a s e t of o b s e r v e d( e x p e r i m e n t a l )d a t ap o i n t s .S u p p o s et h a to n eh y p o t h e s i z e s
t h a t t h e mean v a l u e of the p e r f o r m a n c e v a r i a b l e y of given x is a linear function of
c e r t a i nf u n c t i o n s f.@)
1
of t h ei n d e p e n d e n tv a r i a b l e s xi, i=1, ...,n.Theexpected
value of y is
P

or

n = Bo + Ui
i fi@,

wheredenotesthe mean v a l u eo f y f o rg i v e nv a l u e s of x. Forexample, i f f i&) = Xi

and p = n , t h e n

n = Bo + B1xl,.. . + Bnxn (2-4)

15
 i s a l i n e a rf u n c t i o no ft h e x I f o n et h eo t h e r hand f . & ) = l/xi and p = n ,t h e n
i' 1

rl = Bo + B1/xl + ... + Bn/xn

is n o t a l i n e a r f u n c t i o n o f t h e x 's. However, i t is a l i n e a r f u n c t i o n of t h e B's
i
which are t o b e e s t i m a t e d f r o m t h e g i v e n d a t a by t h e method of l e a s t squares.Thus
t h e estimates bo , bl, ..., bP of B1, ..., BP are g i v e n by t h ev a l u e s of t h e B's which
minimize the sum o f s q u a r e o f d e v i a t i o n s

s = CIy - Bo - CBi fib))

2
.
Certainassumptions are made i n t h i s s o l u t i o n , namely t h a t t h e y = f.(x)aredistri-
i I -
b u t e da b o u tt h ec o r r e s p o n d i n g means
i
TI = B
o
+ I
CB. f . ( ~ )w i t hc o n s t a n tv a r i a n c ea n d
1 -
t h a tt h e y are i n d e p e n d e n to b s e r v a t i o n s . The s o l u t i o n t o t h e l e a s t squaresproblem
i s o b t a i n e d by s o l v i n g a s e t of p + 1 equations in p + 1 unknowns, o f t e n r e f e r r e d t o
as n o r m a le q u a t i o n si nt h el i t e r a t u r e[ R e f . 2-41.
I n many physical problems the model form cannot be expressed as simply as above
( i . e . as a l i n e a rf u n c t i o ni nt h e unknown c o n s t a n t s B
i'
i=O, 1, ..., p ) ,b u t i s non-
l i n e a r ,s u c h as
- BIX
y = Bo(l - e ).
Insuchexamplesiterativeproceduresmustbeusedtosolveforthebest estimates of
t h ec o n s t a n t si nt h e least squaressense.Forexample, see Ref.2-6concerning two
b a s i ca p p r o a c h e s . Computer p r o g r a m sh a v eb e e nw r i t t e nt op e r f o r mt h ei t e r a t i o n .S e e
Ref. 2-7 f o r e x a m p l e .T h i sp r o b l e mr e q u i r e st h eu s e of a g e n e r a lt e c h n i q u ef o r
s o l v i n g a system of n o n l i n e a r e q u a t i o n s , e . g . , t h e Newton-Raphson t e c h n i q u eo ro n e
ofthesearchtechniqueswhichhavebeenwidelyappliedforsuchproblems.
Althoughthe least s q u a r e s c u r v e - f i t t i n g method is most f r e q u e n t l y u s e d , it is
n o ta l w a y st h e , m o s td e s i r a b l e .I n some s i t u a t i o n s o n e w i s h e s t o f i t t h e d a t a b y a
c u r v e which m i n i m i z e s t h e g r e a t e s t d i s t a n c e b e t w e e n t h e f i t t e d c u r v e a n d t h e g i v e n
d a t a ,w h e r e a st h e least s q u a r e s method m i n i m i z e s t h e sum of s q u a r e s of t h e d i s t a n c e s .
Forexample, if the data are p r e c i s e i n t h e s e n s e t h a t t h e y a r e r e s u l t s of some mathe-
m a t i c a lc a l c u l a t i o n( s u c h as t h e s o l u t i o n s o f a differentialequation at a particular
v a l u eo ft h ei n d e p e n d e n tv a r i a b l e ) i t may b e d e s i r a b l e t o r e l a t e thesolution,which
may be a p e r f o r m a n c e m e a s u r e o f i n t e r e s t , t o t h e v a l u e s o f c e r t a i n d e s i g n p a r a m e t e r s
in order to reduce the need for solving the differential equations many times.
As anexample, i nt h ed e s i g no fn u c l e a rr e a c t o r s a problemofimportance to
thedesignengineer i s t h eh o ts p o ti ns a n d w i c h - t y p ef u e le l e m e n t sw h i c hc o n t a i n a

16
uranium a l l o y as t h e c e n t e r s e c t i o n andanotheralloyfortheexternalplates. The
differential equations used in solving for the maximum t e m p e r a t u r e are q u i t e i n v o l v e d
and r e q u i r ec o n s i d e r a b l ec o m p u t i n g time on a modern high-speedcomputer.Consequently,
i t is d e s i r a b l e t o make u s e of s o l u t i o n s of t h e s e e q u a t i o n s f o r s e v e r a l p a r a m e t e r s
toinfer w h a tt h es o l u t i o n i s f o ro t h e rp a r a m e t e rv a l u e s . The s o l u t i o n s t o t h e equa-
t i o n s are e x a c t s u b j e c t t o d i s c r e p a n c y i n t h e model. Thus i t is n o t as meaningful
i n t h i s case b u t t o m i n i m i z e t h e sum of s q u a r e s o f d e v i a t i o n s b e t w e e n t h e f i t t e d
curveandthegivendata as i t i s t o m i n i m i z e t h e l a r g e s t a b s o l u t e d e v i a t i o n b e t w e e n
t h e two. A l i n e a r programming t e c h n i q u e can b e u s e d t o s o l v e t h e p r o b l e m f o r l i n e a r
approximations.
Optimization
The b a s i c p r o b l e m i s : g i v e n y = f w , 5 = (x1,x2, ...,xn ) i n some r e g i o n R, t o
determine the value of 5 that minimizes or maximizes y.
T h i s is a common problem i n a n a l y s i s ; t h e optimum s o l u t i o n i s d e s i r e d , where
optimum is d e f i n e d by means o f a n o b j e c t i v e f u n c t i o n s u c h as c o s t , r e l i a b i l i t y , o r
performance as a f u n c t i o no fs y s t e md e s i g np a r a m e t e r s .I ng e n e r a lt h e x i i = l , . . . ,
n are n o to n l yc o n f i n e dt o some r e g i o n ,b u tp a r t i c u l a rf u n c t i o n so ft h e x i must
s a t i s f yg i v e nd e s i g nc o n s t r a i n t s . Theformof t h eo b j e c t i v ef u n c t i o n and t h a t of
t h ec o n s t r a i n tf u n c t i o nd i c t a t et h et y p eo fp r o c e d u r e ( s )t h a ta p p l y .F o re x a m p l e ,
iftheobjective and t h e c o n s t r a i n t f u n c t i o n s are b o t h l i n e a r , a l i n e a r programming
(LP) approachcanbe made. I ft h eo b j e c t i v ef u n c t i o n i s q u a d r a t i c( n o n l i n e a r ) ,t h e n
a quadratic(nonlinear) programming t e c h n i q u e w i l l b e u s e d i n d e t e r m i n i n g t h e optimum
p a r a m e t e rv a l u e s .I ft h e r e a r e no c o n s t r a i n t s ,s u c h as i nt h ec a s eo ft h e least
s q u a r e se q u a t i o n sf o rn o n l i n e a rm o d e l s ,s e a r c ht e c h n i q u e so rg r a d i e n tt e c h n i q u e s are
used i n most s i t u a t i o n s .S e e Ref. 2-2 f o r a f u r t h e rd i s c u s s i o n of t h e s ep r o c e d u r e s .
The f o l l o w i n g t a b l e c o n t a i n s a l i s t i n g o fo p t i m i z a t i o np r o g r a m sc a t e g o r i z e db y
themathematicalproblem area s u c h as d e s c r i b e d a b o v e . A d d i t i o n a l l i t e r a t u r e r e f e r e n c e s
concerningtheparticularprograms are n o t e d a f t e r t h e program i d e n t i f i c a t i o n number.
The p r e f i x t o t h e number when p r e s e n ti n d i c a t e st h em a c h i n ec o n f i g u r a t i o n .B e c a u s e
a l a r g e number of LP programs are a v a i l a b l e n o a t t e m p t i s made t o g i v e a complete
l i s t i n g of t h e s e . However, f o rt h er e m a i n i n gc a t e g o r i e so fp r o g r a m st h el i s t i n gs h o u l d
be reasonably complete with the exception of programs for dynamic programmingand the
analyticaltechniquesofdifferentialcalculus and c a l c u l u s of v a r i a t i o n s .
Inthecaseof dynamicprogramming i t is o n l y p o s s i b l e t o w r i t e programswhich
solve a particulartypeor class ofproblem,such as a r e l i a b i l i t y o p t i m i z a t i o n problem.
If the problemcanbesolved by methods of d i f f e r e n t i a l c a l c u l u s , t h e n t h e a n a l y t i c a l
problem becomes one of s o l v i n g t h e r e s u l t i n g s y s t e m o f e q u a t i o n s f o r t h e l o c a t i o n of
the stationary points andhenceof testing the nature of the function or the matrix

17
 Table 2-1
Listing of Optimization Programs by Mathematical Programming Problems
OBJECTIVE FUNCTION

CONSTRAINT
FUNCTION Linear r ~~
(2)
Quadratic
~~
(3)
ieparable(Stagewise)
~
(4)
Non-linear-Not (2) or (3)
~~ -
Linear Linear Mratic Dynamic Non-linear
Programming Programming Programming Programming
. Deterministic 7040-H1 3326QPF4 Many programs cited 7040-H9 IBM0007 [Ref.2-81
. Integer in the literature 7090-H9 IBM 0021 [Ref.2-81
. Stochastic
""-""
for specific prob- 7090-H2 3430GPGO [Ref .2-81
lems; see Refs. 7040-H2 3429GP40 [Ref.2-81
7040-CQ-12X [Ref.2-81 2-8 through 2-12. 7040-H2 3189SORT [Refs.2-8
7094-K1 3206M3 [Ref.2-81 and 2-91
7040-H1 3384LSOB [Ref.2-81 7090-H1 3199NLP [Ref.2-81
3600-15.2.001
[Ref.2-81
LIP 1
SHARE (SDA3335) [Ref. 2-10]
IP01,2,3
SHARE(1192,1191 and 1190)
[Ref.2-10]

Nonlinear Non-Linear Calculus of Variations

Programming
7094-K1 3206M3 [Ref.2-81
(See column(4) - Nonlinear
Programming Problems--
Calculus of Variations).

No Differential Search Techniques

Calculus
Constraint 7090-HO 3214MINS [Ref.2-81
0709-C3 3376SEAR [Ref .2-8-1
MINI [Ref. 2-11]
BOTM [Ref.P-ll]
FIBONACCIAN [Ref.2-71
DIRECT SEARCH[Ref. 2-71
ROSENBROCK [Ref.2-121
SCOOP [Ref. 2-11]
I
ofsecondderivatives a t each of the extrema1 p o i n t s t o d e t e r m i n e w h e t h e r i t is a maximum,
a minimum, an i n f l e c t i o n , o r a s a d d l ep o i n t .T h e s ec a l c u l a t i o n sc a n a l l beperformed
n u m e r i c a l l yi fd e s i r e d . No programs are i d e n t i f i e d i n t h i s area. S i m i l a rt r e a t m e n t
of the method o f L a g r a n g i a n m u l t i p l i e r s is p o s s i b l e f o r some p r o b l e m s w i t h c o n s t r a i n t s .
However, i f a l i n e a r , q u a d r a t i c , o r n o n l i n e a r programming t e c h n i q u e is a p p l i c a b l e , t h e
method of Lagrangian multipliers is p r o b a b l y n o t g o i n g t o b e e f f i c i e n t .
Ref. 2-2 c o n t a i n s s u m m a r i e s o f s e v e r a l p u b l i c a t i o n s i n whichone o r moreof the
optimizationproceduresareapplied.
Simulation
The problemstatement is: given a p r o c e s s o r s y s t e m which y i e l d s a n o u t p u t y
f o rg i v e ni n p u t s 5, c h a r a c t e r i z e t h e o u t p u t y. One a p p r o a c h t o d e s c r i b i n g an o u t p u t y
is t o s i m u l a t e t h e p r o c e s s by g e n e r a t i n g t h e i n p u t s xl, ..., xn by a na p p r o p r i a t ep r o -
cedure, such as t h e u s e o f a randomnumber generator,andthenuse a systemmodel to
o b t a i nt h eo u t p u t .T h i sp r o c e d u r e is r e p e a t e d a s u f f i c i e n t number of times t o
characterize the output to the degree of p r e c i s i o n d e s i r e d .
A greatvariety ofproblemscanbesolved by simulation.Forexample, if y = f ( d
is a complex functionof random v a r i a b l e sx l , ..., xn t h e nt h ed i s t r i b u t i o n of t h e
random v a r i a b l e y can be estimated by performing a s u f f i c i e n t number of Monte C a r l o
runs. Sucha procedure i s o f t e n u s e d i n r e l i a b i l i t y a n dp a r a m e t e rv a r i a t i o na n a l y s i s
a s a m e a n s of e s t i m a t i n g t h e p r o b a b i l i t y t h a t t h e p e r f o r m a n c e m e a s u r e of i n t e r e s t
will f a l l i n s i d e c e r t a i n l i m i t s ; such Monte C a r l o t e c h n i q u e s a r e d i s c u s s e d later i n
the report.
Simulationcanbeappliedto random walkproblems,such as t h a t of a n e u t r o n
p a r t i c l e i n a nuclear reactor, to the behavior of a s e q u e n t i a l t e s t proceduregiven
certainassumptionsconcerningtheunderlyingdistributions,ortodiffusionproblems.
An industrial process canbesimulatedforthepurposeofimprovingtheefficiency.
R e p a i ra n ds e r v i c e time (queueing)problems are exampleswhich may r e q u i r e t h e u s e
of s i m u l a t i o nt e c h n i q u e s . Of c o u r s e many of t h e aboveproblems, ifsufficiently
simple, can be treated analytically and t h e u s e of a Monte Carlo procedure is w a s t e f u l .
I n many r e a l w o r l d a p p l i c a t i o n s , however, t h e c o m p l e x i t y is s u c h t h a t t h e u s e of approxi-
mationsor a simulation i s required.
Differentiation
The problem is: given y = f ( d , determine e , a2
1 j
a2 , etc.
,-
axi

Thisproblemcanbetreated by t h e a p p r o p r i a t e c o m b i n a t i o n of t h e t e c h n i q u e s g i v e n
above.However, i t i s a b a s i cp r o b l e mo ff r e q u e n ta p p l i c a t i o n and u s e st h et e c h n i q u e s
of d i f f e r e n c ec a l c u l u s .F o r example,oneobviousprocedure f o ro b t a i n i n gt h ef i r s t
d e r i v a t i v e of a g i v e nf u n c t i o n at point x is toevaluatethefunction at three
1

19
e q u a l l ys p a c e dp o i n t s x
0
, x1 , x2 a n da v e r a g et h ec o r r e s p o n d i n gs l o p e so ft h es e c a n t
l i n e sc o n n e c t i n gt h ep o i n t s as shown i n t h e f o l l o w i n g f i g u r e . Thus t h e estimate
of t h e d e r i v a t i v e is

This i s a c e n t r a l d i f f e r e n c e f o r m u l a ; c l e a r l y many o t h e r s u c h f o r m u l a s c a n b e o b t a i n e d .
S i m i l a r l yo n e can o b t a i n a f o r m u l af o r a mixed o r p u r e s e c o n d p a r t i a l d e r i v a t i v e . F o r
example,Ref. 2-13 c o n t a i n s many suchformulas. I t i s w o r t hn o t i n gt h a tn u m e r i c a l
differentiation,interpolation from a s e t of t a b l e s , and t h e n u m e r i c a l q u a d r a t u r e
formulasused intheconstructionoftablesorforlooking up v a l u e s i n t a b l e s h a v e
much i n common.
Theproblemof differentiation can occur in many ways i n e n g i n e e r i n g a n a l y s i s
problems. We may w i s ht op e r f o r m a sensitivity analysis in which t h e r e l a t i v e c h a n g e s
i nt h ep e r f o r m a n c em e a s u r e s are n e e d e d c o r r e s p o n d i n g t o c h a n g e s i n e a c h of t h e i n d e -
pendentvariables; we may b e s e a r c h i n g f o r an optimumand require the gradient of t h e
function f(x); or w e may wishtoexpand a function in a T a y l o r series t o o b t a i n a
simpleapproximatingfunction.
Integration,Definite and I n d e f i n i t e
X
The problem is: g i v e nt h ef u n c t i o n f ( x ) ,d e t e r m i n eF ( x ) = 1 f (u)du.
a
Sinceintegration is theinverseofdifferentiation,the same b a s i c t e c h n i q u e s ,
a g a i ns t a r t i n gw i t ht h ed i f f e r e n c ee q u a t i o n s , are required.Forexample,thewell

/,}y*-yl
yo
y1-yo
I I I
I I I

X
s2

F i g u r e 2-5. E s t i m a t i o no ft h eD e r i v a t i v e ,d y / d x

20
known t r a p e z o i d a l r u l e is u s e d t o o b t a i n a definite integral of f ( x ) Over an i n t e r v a l
[ a , b ] as shown i n F i g . 2-6.

=
h
-
2
[yo + 2y1 + - .. + 2Yn-1 + Ynl (2-8)

More p r e c i s e f o r m u l a s f o r t h e d e f i n i t e i n t e g r a l c a n b e o b t a i n e d by using second

degreeapproximations(Simpson'srule) and higherdegreepolynomicalapproximations.
Ref. 2-4 c o n t a i n ss e v e r a ls u c hf o r m u l a s .
Inthecase of an i n d e f i n i t e i n t e g r a l and d i f f e r e n t i a l e q u a t i o n s , i t is t y p i c a l
tousethedifferenceformulas and T a y l o r series approximations t o estimate t h e i n t e -
gral function step by s t e p o v e r a given interval starting with known v a l u e s g i v e n
by boundary conditions.

F i g u r e 2-6. N u m e r i c aIl n t e g r a t i o n

21
References

2-1.. Computer Design Publishing Corporation: Computer Industry Annual 1967-1968.

West Concord, Mass., 1967.
2-2. Katz, D. L., et. al.: Computers in Engineering Design Education,Vol. I -
Summary. University of Michigan, College of Engineering, 1, 1966.
April

2-3. Wilde, Douglass J.: Optimum Seeking Methods. Prentice-Hall, Englewood

Cliffs, New Jersey, 1964.
2-4. Pennington, R. H.: Introductiory Computer Methods and Numerical Analysis.
Macmillan, New York, 1965.

2-5. Hamming, R. W.: Numerical Methods for Scientists and Engineers. McGraw-Hill,
New York, New York, 1962.

2-6.. Hartley, H. 6.: The Modified Gauss-Newton Method for the Fitting of Nonlinear
Regression Functions by Least Squares. Technometrics, Vol.
3, no. 2, May
1961, pp. 269-280.

2-7. Nelson, A. C., et. al.: Evaluation of Computer Programs for System Performance
Effectiveness. Progress Report No. 1 (Lab Project 920-72-1, SF-013-14-03,
Task 1604, Contract NO0140 0499),
66C Research Triangle Institute, System
Statistics Group, October 1966.

2-8. Colville, A. R. Mathematical Programming,"Programs Available From IBM Program

Information Department". IBM, Dec., 1966, 20p.

2-9. McCormick, G. P., et al.: Computer Program Implementing the Sequential

Unconstrained Minimization Technique for Nonlinear Programming. AD 621 991,
April 1965, 87p.

2-10. Balinski, M. L.: Integer Programming: Methods, Uses, Computation. Management

Science, Vol. 12, no. 3, Nov. 1965, AD 627 193.

2-11. Shapiro, M. S . ; and Goldstein, Max: A Collection of Mathematical Computer

Techniques. AEC Computing and Applied Mathematics Center, Courant Institute
of Mathematical Sciences, New York University, Feb. 1965, Contract No.
AT (30-1)-1480.

2-12. Rosenbrock, H. H.; and Storey, C.: Computational Techniques for Chemical
Engineers. Pergamon Press, 1966, 328p.

2-13. Abramowitz, M.; and Stegun,I. A.: Handbook of Mathematical Functions with
Formulas, Graphs, and Mathematical Tables. Dover Publications, New York,
New York, 1965.

22
3. Reliabilityandthe Computer -- A Perspective
The s c o p e of a c t i v i t y i n c l u d e d u n d e r t h e h e a d i n g o f r e l i a b i l i t y g e n e r a l l y c a n b e
s u b d i v i d e di n t o two areas: management and c o n t r o lv e r s u sa s s e s s m e n ta n da s s u r a n c e .
The f o r m e r o f t h e s e t y p i c a l l y i n c l u d e s t a s k s s u c h as p l a n n i n g , r e p o r t i n g , t r a i n i n g ,
etc. The r o l e of t h e computer i n t h i s area is mainlyone of bookkeepingandinforma-
tionstorage and r e t r i e v a l . T h e s e u s e s ofcomputers are n o t t r e a t e d i n t h i s r e p o r t .
As a real a i d t o r e l i a b i l i t y , t h e c o m p u t e r ' s most v i t a l f u n c t i o n i s i n performing
complex d a t a p r o c e s s i n g and a n a l y s i s o p e r a t i o n s w h i c h p r e v a i l m o s t l y i n t h e a s s e s s m e n t
a n da s s u r a n c ea c t i v i t i e s .T h e s er o l e s are theonesemphasized i nt h i sr e p o r t . The
major tasks i n w h i c h c o m p u t e r s c a n a i d r e l i a b i l i t y w i t h t h e s e f u n c t i o n s are i d e n t i f i e d
below, then surveyed for a perspectiveontherolethatcomputerscanplayinimple-
menting them.
F a i l u r e modes and e f f e c t s a n a l y s e s (FMEA) are p r o c e d u r e s f o r c o n s i d e r i n g modes
of operationofcomponents(such as a s h o r t o f a resistororprematureoperation
of a t r a n s m i t t e r ) and t h e e f f e c t s t h e s e modes have on systemoperation.Parameter
v a r i a t i o na n a l y s e s (PVA) t r e a t v a r i a t i o n s i n p e r f o r m a n c e u s i n g models ( e i t h e r mathe-
matical o r p h y s i c a l ) which relate p e r f o r m a n c e t o c h a r a c t e r i s t i c s of t h e components
and o p e r a t i n gc o n d i t i o n st h a tc a u s et h ep e r f o r m a n c et ov a r y . P a r t a p p l i c a t i o na n a l y s e s
consider individually the parts and componentsof thesystemfor a comparisonof
o p e r a t i n gc o n d i t i o n st or a t e dc a p a b i l i t i e s .R e l i a b i l i t yp r e d i c t i o n is concernedwith
the probability of s u c c e s s f u l o p e r a t i o n o f a n e q u i p m e n t u s i n g m o d e l s t h a t r e l a t e system
successprobabilitiesofeventsassociatedwith componentsand o p e r a t i n gc o n d i t i o n s ;
i t c a ni n c l u d ep r o b a b i l i t i e sr e l a t e dt ob o t hl i f e and performance.Testing is con-
cernedwith a l l effectsintroducedabove; i t alonecanbe a means t o a n end o r s e r v e
both a supplementary andcomplementary roletotheanalyses by s u p p l y i n gi n f o r m a t i o n
t os u p p o r tt h ef o r m u l a t i o no fm o d e l s ,d a t ai n p u t st o them,and c h e c k so ft h e i rv a l i d i t y .
The f i r s tf o u r o ft h e s e are a n a l y s i st a s k si n i t i a t e de a r l yi nd e s i g n . A
perspectivefortheircoordinatedimplementationfortreatingreliabilityproblems
indesign anddevelopment is i l l u s t r a t e d i n F i g . 3-1, w h i c h a l s o i n c l u d e s a general
i n d i c a t i o n ofcomputer u t i l i t yf o rp e r f o r m i n gt h e s et a s k s .T e s t i n ga l s oo f t e n employs
computermethodsand as n o t e d earlier, t h i s t a s k serves as s u p p o r t t o t h e a n a l y s e s .
The p r o p o s e dd e s i g na n dm i s s i o nd e f i n et h ep r o b l e mt ob ea n a l y z e d . The analysespro-
v i d et h eo u t p u ti n f o r m a t i o nf o rd e s i g n improvementand assurance. Improvement r e s u l t s
through a f e e d b a c kp r o c e s sw h e r e b yt h ed e s i g no rm i s s i o n is modified as required.Such
modificationsrequiretradeoffsbetweenreliability and o t h e r r e q u i r e m e n t s of t h e
s y s t e m( c o s t ,m a i n t a i n a b i l i t y ,e t c . )b e f o r eb e i n g made.
In brief, the overall objectives of t h e r e l i a b i l i t y a n a l y s e s are:

23
 I Reliability
Prediction I

I Failure /
I
Modes / I For
Design I ?aramc ter I Design

,+
and /
and Jari: t i o n + I Improvement
L*3 Mission
Effects
Analysis
l n a l y si s
/
I and
A
I * I Assurance
I / I
\ /
\
\ /
\ / *Computer methodscanbeused.
\ /
**Computer methods are o f t e n u s e d .
*** Computer methods u s u a l l y s h o u l d
Application
1 Anaipis I beused.

Figure 3-1. R e l i a b i l i t yA n a l y s e si n DesignandDevelopment

 ( 1 )i d e n t i f y i n ga n dr e m o v i n gp o s s i b l ec a u s e so ff a i l u r e ,
(2) b a l a n c i n gs a f e t y( d e s i g n )m a r g i n so ra p p o r t i o n i n gt o l e r a n c e s ,a n d
(3) o b t a i n i n gn u m e r i c a la s s e s s m e n t so fr e l i a b i l i t y .
None o f t h e d e f i n e d r e l i a b i l i t y t a s k s is c a p a b l e o f a c h i e v i n g t h e s e o b j e c t i v e s by
itself. A s i l l u s t r a t e d ,t h et a s k s are s t r o n g l yi n t e r r e l a t e d ; i t is t h r o u g ht h e i r
coordinatedapplicationandthe combined u s e o f t h e i r r e s u l t s t h a t maximum b e n e f i t
is d e r i v e d f o r r e l i a b i l i t y . The computercan a i di n 'p e r f o r m i n ge a c ho ft h ei n d i v i d u a l
tasks;forexample, it u s u a l l y s h o u l d b e u s e d f o r PVA a n d o f t e n s h o u l d b e f o r relia-
bilityprediction. Eachof t h et a s k s and therelevanceofcomputermethodsto im-
plementing each are d i s c u s s e d below.
F a i l u r e modes and e f f e c t s a n a l y s i s s e r v e s t h e p u r p o s e o f r e v e a l i n g whatcan
happen t o t h e s y s t e m . By c o n s i d e r i n g t h e l i k e l i h o o d a n d t h e c r i t i c a l i t y o f t h e p o s s i -
b l e modes of s y s t e m b e h a v i o r , i t allowsdirection of e f f o r t i n t h e o t h e r r e l i a b i l i t y
tasks. It d e f i n e ss p e c i f i c modes o fb e h a v i o rf o rp e r f o r m a n c ev a r i a t i o ns t u d i e s ; it
identifies critical areas t o b e e m p h a s i z e d i n p a r t a p p l i c a t i o n a n a l y s e s ; it designates
f a i l e d s t a t e s t ob ei n c l u d e di nr e l i a b i l i t yp r e d i c t i o n s .B e c a u s eo f its valuein
directingothereffort, a f a i l u r e modes and e f f e c t s a n a l y s i s s h o u l d b e i n i t i a t e d
earlyinthedesign program. A computer is seldomused inidentifyingfailure modes;
i t is used i n i n v e s t i g a t i n g f a i l u r e e f f e c t s as d i s c u s s e d i n S e c . 6 of t h i s r e p o r t .
Parameter v a r i a t i o n a n a l y s i s i s concernedwiththeassurancethatperformance
is a c c e p t a b l e . Whereas r e l i a b i l i t y p r e d i c t i o n , f a i l u r e modes and e f f e c t s a n a l y s i s , and
part application analysis are u s u a l l y f o r m a l t a s k s i n s y s t e m c o n t r a c t o r a c t i v i t i e s
p a r a m e t e rv a r i a t i o na n a l y s i sh a sb e e nn e g l e c t e dd u et ol i m i t e du n d e r s t a n d i n go f
theavailabletechniquesfortreatingperformancevariability.
A s d e s c r i b e d i n S e c . 4 a number ofanalyticaltechniqueshavebeenassembled
and t e s t e d , and a f l e x i b l e PVA programhasbeenwritten.Inthisprogram,mathematical
orphysical models are used t o r e l a t e p e r f o r m a n c e a t t r i b u t e s t o componentand interface
c h a r a c t e r i s t i c s .P r o b a b i l i s t i ct e c h n i q u e ss u c h as p r o p a g a t i o no f moments and Monte
Carlosimulation are u s e d t o estimate p r o b a b i l i t i e s o r d i s t r i b u t i o n s o f p e r f o r m a n c e .
Various end-limit techniques provide worst-case performance values and parametersen-
s i t i v i t i e s .S o u r c e s of v a r i a t i o n are i d e n t i f i e d and r e l a t i v ec o n t r i b u t i o n so f com-
p o n e n tv a r i a t i o nc a nb ed e t e r m i n e d .p a r a m e t e rv a r i a t i o na n a l y s e sy i e l dd i r e c t l y
u s e f u ld e s i g ni n f o r m a t i o na n d , as i l l u s t r a t e d i n F i g . 3-1, provideinputstothe
o t h e rt a s k s .T h e s ei n c l u d e ,f o re x a m p l e ,o p e r a t i n gc o n d i t i o n sf o rc o m p o n e n t su s e d
in application analyses and performance estimates t o b e i n c l u d e d i n r e l i a b i l i t y
predictions.
P a r t a p p l i c a t i o n analysis determineswhethercomponents are p r o p e r l y a p p l i e d .
F o re x a m p l e ,t h e r m a la n de l e c t r i c a ll o a d s on p a r t s are u s e d f o r a p p r o p r i a t e a d j u s t m e n t

25
of f a i l u r e rate estimates, and p a r t s w i t h l o a d s e x c e e d i n g d e s i g n s p e c i f i c a t i o n s are
a p p r p p r i a t e l yr e s p e c i f i e do rt h ed e s i g n changed t or e d u c et h e s el o a d s .C o m p u t e r s
are r e a d i l y u s e d t o make a p a r t a p p l i c a t i o n a n a l y s i s ; s u c h a n a n a l y s i s is frequently
conducted as a p a r t of a l a r g e ra n a l y s i s .F o re x a m p l e , i t i s e a s y when performing
a circuitanalysistocheckactualvoltage,current, and power a g a i n s t r a t e d v a l u e s
foreach component i n t h e c i r c u i t , a n dp r o v i s i o n sf o rd o i n gt h i s are i n c o r p o r a t e d
i n some c i r c u i t a n a l y s i s p r o g r a m s .F u r t h e rd i s c u s s i o n i s g i v e ni nS e c . 5.
Reliabilitypredictions are based on l o g i c r e l a t i o n s h i p s e x p r e s s i n g s u c c e s s o r
f a i l u r ee v e n tp r o b a b i l i t i e s of systemcomponents.Currently,mostprediction cal-
c u l a t i o n s are b a s e do nt w o - s t a t e( s u c c e s sv s .f a i l u r e )m o d e l su s i n gp a r tf a i l u r e
rates andexponential lifedistributions. Becauseof t h e many s i m p l i f y i n ga s s u m p t i o n s ,
l i t t l e s i g n i f i c a n c ec a nb ea t t a c h e dt ot h em a g n i t u d e s of t h e numbersobtained. Some
advancedtechniquesconsider more t h a n two s t a t e s as d i s c e r n e d by t h e f a i l u r e modes
and e f f e c t s a n a l y s i s , and more a p p r o p r i a t e l i f e d i s t r i b u t i o n s are a l s o a v a i l a b l e .
Althoughtheframeworkhasbeendevelopedforincludingperformancedegradation
failuresinprediction,thevalue of r e l i a b i l i t y p r e d i c t i o n a t p r e s e n t l i e s more i n
thedesignweaknessesdetectedinperformingtheanalysis and t o compare a l t e r n a t i v e
d e s i g n st h a ni nt h ea c t u a l numbers r e s u l t i n g . The a p p l i c a t i o n of t h e s et e c h n i q u e s
bycomputers is t r e a t e d i n S e c . 7. Computers p l a y a v a l u a b l e r o l e by e n a b l i n g more
realistic prediction models to be employedand by performing the computations which
producethereliability estimates r e s u l t i n g f r o m t h e s e m o d e l s .
Each methodabove separatelyprovidesusefuldesigninformation,buttoassure
appropriateemphasisonbothperformance and l i f e , t h e r e s u l t s f r o mt h ev a r i o u s
methodsmustbeconsideredjointly.Becauseofthedifferentforms of t h e r e s u l t s
t h ec o m b i n a t i o np r o c e s s is p r i m a r i l ys u b j e c t i v e , so thecomputercanprovide little
h e l ph e r e . As anexample on t h ec o m b i n a t i o n of t h et a s k s ,s u p p o s et h a t parameter
variationanalyseshaveyieldedworst-caseresultsfor two d e s i g n s b e i n g compared
and t h a tD e s i g n A h a s smaller v a r i a t i o n st h a nD e s i g n B. R e l i a b i l i t yp r e d i c t i o n sw i t h
c o n v e n t i o n a lt w o - s t a t ea n a l y s e s may, i n t u r n , i n d i c a t e t h a t D e s i g n B has a higher
p r o b a b i l i t yo fs u c c e s s .I n d i c a t i o n s are t h u st h a tD e s i g n B r e p r e s e n t sa n improvement
i n l i f e overDesign A, however a t a s a c r i f i c eo fp e r f o r m a n c e .I ft h e r e is a d e q u a t e
confidenceintheresultsofeach, a t r a d e - o f f may benecessary,forexample, re-
s u l t i n g i n Design C t h a t u s e s some o f t h e b e t t e r f e a t u r e s o f D e s i g n s A and B. On
the other hand, l a c k of c o n f i d e n c e i n t h e r e s u l t s may d i c t a t e t h e need f o r more
s o p h i s t i c a t i o ni nt h ea n a l y s e s .F o re x a m p l e ,a ne x t e n s i o no fp r e d i c t i o nt o more
realistically include additional modes of p a r t f a i l u r e s and t h e i r e f f e c t s may show
t h a t Design A is t h e b e t t e r from t h e s t a n d p o i n t o f l i f e .

26
 No o n e o f t h e r e l i a b i l i t y t a s k s p r o v i d e s a "cure-all" for reliability,but through
their coordinated andcombined use, the maximum a s s u r a n c e f o r r e l i a b i l i t y is achieved.
Alsoathe responsibility for reliability cannot be delegated to reliability specialists
a l o n e .R e l i a b i l i t y is a r e s p o n s i b i l i t y of a l l p e r s o n n e l , b u t t h e m a j o r r e s p o n s i b i l i t y
rests w i t h t h e d e s i g n e r . Good e n g i n e e r i n g is, and w i l l r e m a i n ,t h em a j o rk e yt o re-
liability. Themethods are provided as a s u p p l e m e n tt o ,b u tn o t a substitutefor,
good e n g i n e e r i n g p r a c t i c e .
J u s t as p e r f o r m i n g t h e s e t a s k s is no s u b s t i t u t e f o r good e n g i n e e r i n g , n e i t h e r
is theindiscriminateuseofthecomputertoperformsuchtasks good r e l i a b i l i t y
engineering. Computer m e t h o d s s h o u l d b e s e l e c t i v e l y u s e d i n d e s i g n f o r r e l i a b i l i t y ,
and usedonly when t h e y c a n p r o v i d e g e n u i n e l y u s e f u l r e s u l t s w i t h i n t h e e c o n o m i c , time
a n do t h e rr e l e v a n tc o n s t r a i n t s on thedesignunderconsideration.Withinthebounds
of t h e s e c o n s t r a i n t s , t h e computer a i d s t o d e s i g n f o r r e l i a b i l i t y which are d i s c u s s e d
intheremainder of thisreportcomprise a powerful s e t o f t o o l s f o r i n s u r i n g t h a t
a reliableproduct is produced.

27
4. Parameter V a r i a t i o nA n a l y s i s
There are two ways i n which a p i e c e of equipment o r a system can f a i l t o perform
i t s i n t e n d e df u n c t i o n . One is c a t a s t r o p h i cf a i l u r e , w h i c h i s l i k e l yt ob ea b r u p t and
t o have a d r a m a t i c e f f e c t onequipment o rs y s t e mo p e r a t i o n .I na ne l e c t r o n i cc i r c u i t ,
a typical catastrophic failure is theopeningortheshortingof a diode or a transistor.
The o t h e r t y p e o f f a i l u r e is driftfailure,whereduetothevariationsofequipment
p a r a m e t e r sw i t h t i m e , theperformanceoftheequipment a t some t i m e becomesno longer
satisfactory. The p r e d i c t i o n of d r i f t - t y p e f a i l u r e s r e q u i r e s a s t u d y of combinations
of component p a r a m e t e r v a l u e s a n d t h e r e s u l t i n g e f f e c t s o f t h e d r i f t i n g o f t h e s e
v a l u e s on e q u i p m e n tp e r f o r m a n c e .S t u d i e so fp a r a m e t e rd r i f t sa n dt h e i re f f e c t s on
s y s t e mp e r f o r m a n c ec o m p r i s ep a r a m e t e rv a r i a t i o na n a l y s i s (PVA). The a v a i l a b i l i t y o f
thehighspeeddigitalcomputerhas made p o s s i b l e a d r a m a t i c i n c r e a s e i n t h e a b i l i t y
t o p e r f o r me x t e n s i v e PVA s t u d i e s and as a r e s u l t i m p r o v e t h e r e l i a b i l i t y of t h e e q u i p -
ment by m i n i m i z i n g v i a d e s i g n m o d i f i c a t i o n s t h e l i k e l i h o o d o f a drift-type failure.
4.1 PVA Modeling
A PVA model must be adequately accurate to simulate the equipment behavior over
t h ee n t i r er a n g eo fe n v i r o n m e n t se x p e c t e df o rt h ee q u i p m e n t
*. To e n a b l et h e PVA
a n a l y s i st ob ea c c o m p l i s h e d ,t h e modelmust expresstherelationships between t h e
performancecharacteristicsofinterest and a l l t h e p a r a m e t e r s t o b e i n c l u d e d f o r
s t u d y .I n many c a s e s ,t h ee q u i p m e n tp a s s e st h r o u g hs e v e r a ld i s t i n c to p e r a t i n gr e g i o n s ,
and i t i s n e c e s s a r y t h a t t h e modeladequatelyrepresenteachregion. A changefrom
t h e ON t o t h e OFF s t a t e of a t r a n s i s t o r , f o r example,requires a new e q u i v a l e n t c i r c u i t
orthetransistor, a n de a c hs u c he q u i v a l e n tc i r c u i tm u s ta d e q u a t e l ys i m u l a t et h e
actualcircuitoperationtoprovideengineeringconfidenceintheperformance i t pre-
predicts.
A t t h ec o r eo fa n yp a r a m e t e rv a r i a t i o na n a l y s i s i s a mathematicalmodel; in
e x p l i c i t form,

or implicit form

where
Y. ( t ) i s t h e j t h p e r f o r m a n c ea t t r i b u t eo rm e a s u r e ,
J

*Some v e r y p r a c t i c a l v i e w p o i n t s onmodeling are p r e s e n t e d i n Vol. I - Parameter

\
VariationAnalysesofthis series.

28
 X(t) i s a vector comprised of the environment inputs, such as environ-
mental stressesand loads, plus the component characteristics,
t is the time variable, and
g
j'
..
j=l,. ,N is the set of models corresponding to the number
of responses or
the orderof the differential equations which describe the transient
behavior of the system.

For example, the model may be of the form of a system of differential equations,

-
2
ayl -
a:+clat
at
+ c2Y2 - c3

-
ay2
+ C Y + C Y -
at 4 2 51-'6'

where the c depend on the input vector through a set of explicit expressions.
i
The time behavior for the model may appear in of several
one ways. For example,
it may be a gradual deterioration of a component and hence result in a corresponding
change in the values of one or more of the component characteristics.
In order to
analyze an element or system for this type of degradation, the wearout characteristics
of the system must be known or estimates
be must
available.
A second way in which time may appear is through the mission profile. For
example, if itis known that the temperature profile is critical and how the part
characteristics vary with temperature, then
an analysis can be performed
by describing
the temperature-part characteristic behavior by deterministic and/or random processes
and performing the analysis at several times in the mission life.
Time may enter the analysis directly through the transient behavior.
In this
case a program for solving differential equations may be required for relating the
transient characteristics to the pertinent element parameters, inputs, etc.
In
whatever manner time enters the analysis, it is assumed that it may be included by
a procedure such as one of the following:
(1) A deterministic functionof time such as a linear or exponential decay
function.
(2) An autoregressive scheme such as

(3) A stochastic process such as a normal stationary process superimposed on

a deterministic drift.

29
 (4) A system of differential equations.

4.2 Analysis Techniques

Several analysis techniques
are used orPVA on a computer. One of the most
widely used is worst-case analysis.
The worst-case method is a nonstatistical approach
which is intended to determine whether
it is-possible,within the specified tolerance
limits on parameters, for the system performance to fall outside the specified pe
formance limits. The answer is obtained by using performance models, and setting
the parameter values at combinations of upper and lower tolerance limits to obtai
the worst-case performance. A related.PVAtechnique is sensitivity analysis. Worst-
case and sensitivity analys,es typically use the same mathematical techniques, as is
discussed later. The purpose of a sensitivity analysis is to determine how sensitive
a system performance variable
is to variations in input variables.
Another common technique for performing PVA is the moments method. This techniq
combines statistics and system analysis to determine the probability that performan
will remain within the specified limits; the technique is often called the prop
of distributions
of-variance method as second moments are usually the highest moments
used. The method applies the propagation-of-variance formula to the first two moments
of the component part probability density functions to obtain the equivalent moments
of the performance distribution.
The convolution method for
PVA is another approach to obtaining statistical
distributions of output variables. Although potentially a quite general method, the
technique reported here and as implemented by computer programs is a simplified
version of the general convolution method.
In the Monte Carlo method component values are selected randomly; the performa
of each randomly generated configuration of the equipment under study is calculated
and compared with performance specification limits. This technique has the advantage
that any component parameter distribution can be handled; it has the disadvantages
that it requires a lot of computer time and offers little help in identifying and
correcting failures.
The implementation on a computer of of
each
the above techniques is now treated
in detail. Some of the computer programs which are available for implementing the
techniques are discussed.
4.2.1 Worst-case Analysis
The theory on which worst-case analysis is based derives from expressing the
model performance parameters & = (X1 X 2
Y as functions of the input: vector
j
Xm1 ---
and expanding these functions in Taylor series. The input vector consists of all
pertinent part characteristics, inputs, loads, and environment factors. Let the model
for an arbitary performance Y be
parameter

30
 Y = g(xJ = g(X1, x2,. .., Xm). (4-2)

A T a y l o r series e x p a n s i o n a b o u t t h e n o m i n a l v a l u e of Y f o r its change from

n o m i n a l v a l u e is

AY= 1
i=l
ay
- AXi
1
+- 1 7
i=laxi
a2y
(AX,)
2
+ ... , (4-3)

% %
where
AY = change i n v a l u e o f Y from i t s nominal value,
AXi = Xi - X , t h ew o r s t - c a s ed e v i a t i o no ft h ei - t hi n d e p e n d e n tv a r i a b l e
iN X. from i t s n o m i n avl a l u e
1
X
iN , and
&= (X , ..., X ) , t h en o m i n a lv a l u e s of t h e X's.
5* x2N "N
Eq. (4-3) i s a s i m p l i f i e de x p a n s i o nw h i c hi n c l u d e s no c r o s s - p r o d u c t terms; a completely
generalTaylor series expansion i s g i v e n i n Appendix B , Vol. I - P a r a m e t e rV a r i a t i o n
A n a l y s i s of t h i s series. In p r a c t i c et h ec r o s s - p r o d u c t terms are seldomusedeven in
computerprograms,so Eq. (4-3) is t h ee x p a n s i o nm o s tl i k e l yt ob ef o u n d .F r e q u e n t l yo n l y
the linear terms are u s e d ; t h e e x p a n s i o n t h e n h a s t h e f a m i l i a r f o r m

AY =
%
- AX1 + ay AX2 + ...
axl ax2
(4-4)
+ -aY AXm .

To perform a w o r s t - c a s e a n a l y s i s , t h e p a r t i a l d e r i v a t i v e s o f Y withrespect to
e a c hi n d e p e n d e n tv a r i a b l e Xi mustbecomputed.Severaltechniques are used t o compute
d e r i v a t i v e s on a computer. The " E i g h tP o i n tC e n t r a lD e r i v a t i v eF o r m u l a " is a p o p u l a r
method [Refs. 4-1and 4-21. T h i sf o r m u l af o rt h ef i r s tp a r t i a ld e r i v a t i v e is

= a y =- 4
t
('+lh - '-1h) - (+
' 2' - '-2h) + 10 5 ('+3h - '-3h)

(4-5)
--1
280 ('+4h - '-4h)
This formula i s e v a l u a t e d by s t e p p i n g t h e i n p u t p a r a m e t e r Xi f o u r e q u a l i n c r e m e n t s
h each way from i t s n o m i n a l v a l u e Xi , and calculating the value of Y for each step
w h i l eh o l d i n g a l l o t h e ri n d e p e n d e n tN v a r i a b l e s Xj ,j i, a t t h e i rn o m i n a lv a l u e s .

31
 I

Inthisformula, h is e x p r e s s e d as t h e f r a c t i o n a l c h a n g e i n Xi; i f onepercentof

is thestepsize, h i s 0.01.The v a l u e so f Y are t h e n s u b s t i t u t e d i n t o Eq. (4-5)
thepartialderivative aY/aXi o b t a i n e d .T h i s method i s u s e di nt h ew o r s t - c a s e
a n a l y s i s method c a l l e d MANDEX,
* which is perhapsthemostwidelyusedworst-case
computermethod[Ref. 4-21.
A second formula for computing partial derivatives via computer is t h e f i v e - p o i n t
c e n t r a ld i f f e r e n c ef o r m u l a[ R e f . 4-31, The f i r s td e r i v a t i v ef o r m u l a is

1 - ay %
=- 1
'i - axi 12h ('-2h - 8Y-lh + 8+lh - '+2h) *

%
T h i s i s e v a l u a t e da n a l o g o u s l yt ot h ee i g h t - p o i n to n e ,b u t i t s accuracy i s somewhat
less. However, i t s a c c u r a c yu s u a l l y is a d e q u a t e when o n l y t h e f i r s t d e r i v a t i v e s are
used i n t h e T a y l o r series expansion. A f i v e - p o i n tf o r m u l af o rt h es e c o n dp a r t i a l
derivative with respect to one independent variable is

t h e s ef i v e - p o i n te q u a t i o n s are used i n o n eo ft h e PVA programdiscussed later. Eqs.

( 4 - 6 ) and (4-7) are d e r i v e d i n AbramowitzandStegun[Ref. 4 - 4 1 ; n o t et h a ti nt h e s e
e q u a t i o n s h is j u s t a number, n o t a f r a c t i o n of X .
iN
Having e v a l u a t e d t h e p a r t i a l d e r i v a t i v e s , t h e w o r s t - c a s e limits are n e x t computed.
The s i g n s o f t h e f i r s t p a r t i a l d e r i v a t i v e s are examined t o e n a b l e a p r o c e d u r e f o r
computingworst-case limits whichreducescomputing time. A w o r s t - c a s e maximum by
d e f i n i t i o n o c c u r s when theperformanceparameter Y t a k e s on i t s g r e a t e s t v a l u e , i.e.,
when AY is maximum and p o s i t i v e .C o n s e q u e n t l y , a l l i n p u tv a r i a b l e sw i t hp o s i t i v e
f i r s t p a r t i a l s are set a t theirupper limits and a l l w i t h n e g a t i v e f i r s t p a r t i a l s
a t t h e i rl o w e r limits. T h i sp r o c e d u r eg i v e st h ew o r s t - c a s e maximum i n t h e l i n e a r
T a y l o r series expansion, Eq. ( 4 - 4 ) , s i n c ee a c h term is a p r o d u c t of e i t h e r two
positive or two n e g a t i v e q u a n t i t i e s . F o r t h e w o r s t - c a s e minimum, lower limits a r e
used for the X. w i t h p o s i t i v e p a r t i a l s and upper limits f o r t h o s e w i t h n e g a t i v e p a r t i a l s ,
1
producing a l l n e g a t i v e terms and hence the worst-case minimum i n t h e l i n e a r series
summation f o r AY.
It is possible that the partial derivative of an output variable Y withrespect
t oa ni n p u tv a r i a b l e X is n o tl i n e a r ;t h e nt h ea b o v ep r o c e d u r ed o e sn o tn e c e s s a r i l y
i

*MANDEX is an acronym for modified and expanded worst-case analysis.

32
p r o d u c et r u ew o r s t - c a s e limits. L i n e a r i t yc h e c k so r morecomplex series expansions
c a nb ei n c o r p o r a t e d . t op r e v e n ts u c hi n a c c u r a c i e sf r o mg o i n gu n n o t i c e d .T h e s e safe-
g u a r d s are d i s c u s s e d i n S e c . 4.2.5 as t o how t h e y are implemented i n s p e c i f i c PVA
programs.
W o r s t - c a s ea n a l y s i s i s appliedmostwidelytoelectroniccircuits,but it is
equally applicable to any system for which a performance model can be derived and
i n p u tp a r a m e t e rv a r i a t i o n s are known o rc a nb er e a s o n a b l ye s t i m a t e d . The p r o p e r
u s eo fw o r s t - c a s ea n a l y s i s is as a f i r s t s t e p i n t h e PVA s t u d yo f a system. Ifthe
s y s t e mp a s s e st h i sp a r a m e t e rv a r i a t i o na n a l y s i s , i t i s a l m o s tc e r t a i nt op a s sa n y
other. Hence i t i s p o s s i b l e t o a c c e p t a d e s i g ni f i t p a s s e sw o r s t - c a s ea n a l y s i s .
Conversely, i t u s u a l l y is i n e r r o r t o reject t h e d e s i g n o n l y b e c a u s e it f a i l s a
portionof a worst-caseanalysis,sincetheprobability of o b t a i n i n g a t r u e w o r s t -
caseconditioninpractice i s v e r y small. A failuretopass a w o r s t - c a s ea n a l y s i s
usuallyindicatesthatotheranalysesshouldbeperformed.
4.2.2 S e n s i t i v i t yA n a l y s i s
An i m p o r t a n t PVA t e c h n i q u e r e l a t e d t o w o r s t - c a s e a n a l y s i s is a n a l y s i s o f t h e
s e n s i t i v i t y of s y s t e mp e r f o r m a n c et ov a r i a t i o n si ni n p u tp a r a m e t e r s .A l t h o u g hs e v e r a l
differentdefinitions of s e n s i t i v i t y a r e found i n t h e l i t e r a t u r e [ R e f s . 4-3 and 4-51,
in essence the sensitivity of a system i s simply a measure of the effect of parameter
v a r i a t i o n s on t h es y s t e mp e r f o r m a n c e .I ne q u a t i o nf o r ms e n s i t i v i t yc a nb ee x p r e s s e d

Y.
sxi = AY./AX~ ,
J

where
S j i s t h es e n s i t i v i t y of theperformancemeasure Y t ot h ev a r i a t i o ni nt h e
xi j
system model parameter X
i
AY is t hceh a n gien Y and
j j
AXi is t h ev a r i a t i o ni n X
i

An a l t e r n a t i v e form i s t h e n o r m a l i z e d s e n s i t i v i t y

i t i s more f r e q u e n t l y u s e d .
Each o ft h e terms on t h e r i g h t s i d e o f Eq. (4-9) is e i t h e r a v a i l a b l e o r e a s i l y
obtainedfromtheperformancemodel. A l l t h a t i s r e q u i r e dt oo b t a i ns e n s i t i v i t y is
t oc a l c u l a t e AY ( t h ec h a n g ei n Y produced by t h ec h a n g ei n X o n l y ) and thenperform
j j i

33
t h e t h r e e arithmetic o p e r a t i o n s i n d i c a t e d i n Eq. (4-9) f o re a c hp e r f o r m a n c ev a r i a b l e
.Y and i n p u tv a r i a b l e
Xi. T h ec o m b i n a t i o no fw o r s t - c a s ea n ds e n s i t i v i t yi n f o r m a t i o n
j
on a d e s i g n i s complementary, p a r t i c u l a r l y when d e s i g n m o d i f i c a t i o n s are r e q u i r e d .
Suppose a d e s i g n f a i l s t o p a s s a worst-case analysis for a Performancemeasure Y
Y j
w i t hr e s p e c tt o a variable X I fa l s ot h es e n s i t i v i t y S j i s h i g h , e. g. , a 1%
i'
xi
change i n X. produces a 5% change i n Y a r e d e s i g na r o u n dt h ev a r i a b l e X may b e
1 j' i
needed. I fw o r s t - c a s ea n a l y s i sw i t hr e s p e c tt o X f a i l sf o rs e v e r a lo u t p u tv a r i a b l e s
i
Y: and t h e c o r r e s p o n d i n g s e n s i t i v i t i e s a r e h i g h , s u c h a redesign probably is r e q u i r e d .
J
Theaccuracy of a s e n s i t i v i t y c a l c u l a t e d w i t h Eq. (4-9) i s o b v i o u s l yl i m i t e d
by theaccuracyoftheassumptions and approximationsused i nt h ec a l c u l a t i o n .F o r
example, maximum s e n s i t i v i t y may o c c u r somewhere b e t w e e n , r a t h e r t h a n a t , t h eu p p e r
and l o w e ri n p u tp a r a m e t e r limits. Theremarks made f o rw o r s t - c a s ea n a l y s i s on
l i n e a r i t y a n dh i g h e ro r d e r series e x p a n s i o n s a l s o a p p l y h e r e .
4.2.3 Moments A n a l y s i s
The moments methodof PVA a n a l y s i s h a s t h i s name b e c a u s e i t makes u s e o f t h e
moments of t h e s t a t i s t i c a l d i s t r i b u t i o n s of i n p u t p a r a m e t e r s t o o b t a i n t h e moments
of t h ed i s t r i b u t i o n so ft h es y s t e mp e r f o r m a n c em e a s u r e s . As u s u a l l y implemented on
a computer, i t makes u s e o f t h e f i r s t moment ( t h e mean) and thesecond moment about
t h e mean ( t h e v a r i a n c e ) o f t h e d i s t r i b u t i o n s of t h e i n p u t p a r a m e t e r s t o o b t a i n t h e
mean and t h e v a r i a n c e o f t h e d i s t r i b u t i o n s of thesystemperformancemeasures. When
a d i s t r i b u t i o n is n o r m a lt h e s e two moments d e s c r i b e i t c o m p l e t e l y .A l t h o u g hd i s t r i -
butionswhich are found i n p r a c t i c e are s e l d o m p r e c i s e l y n o r m a l , t h e a c c u r a c y is o f t e n
a d e q u a t ef o r PVA p u r p o s e s .T h i ss i m p l i f i e d formof t h e moments m e t h o d ,c a l l e dt h e
propagation-of-variancemethod, i s what i s describedbelow.
The mean v a l u e s f o r t h e model o u t p u t p a r a m e t e r s are o b t a i n e d by programming t h e
computer t o i n s e r t mean v a l u e s f o r a l l the variables in the system model i n p u t v e c t o r
a n dt h e ns o l v et h ep e r f o r m a n c ee q u a t i o n s . Thecomputer t h e nc a l c u l a t e st h es e c o n d
moment a b o u t t h e mean, i . e . , thevariance,ofeachoutputvariable by e v a l u a t i n g t h e
propagation-of-varianceformulagivenbelow. An a d d i t i o n a lf e a t u r ei n c o r p o r a t e di n
some programs i s t h a t e a c h o f t h e terms i n t h e p r o p a g a t i o n - o f - v a r i a n c e f o r m u l a is
d i v i d e d by t h e t o t a l v a r i a n c e t o g i v e a n i n d i c a t i o n o f t h e f r a c t i o n of t h e v a r i a n c e
c o n t r i b u t e d by each input parameter.
The propagation-of-varianceformula is theheartofthe computer-implemented
moments methodof a n a l y s i s .T h i sf o r m u l a i s t h em a t h e m a t i c a ls t a t e m e n tt h a tt h e
performancevariability is the net result of t h e v a r i a b i l i t y of a l l t h e i n p u t p a r a -
meters i n t h e s y s t e m , and t h a tt h ec o n t r i b u t i o no fe a c hi n p u tp a r a m e t e rd e p e n d s upon
its individual variability and on t h e r e l a t i v e i m p o r t a n c e o f t h a t p a r a m e t e r i n

34


d e t e r m i n i n gt h ep e r f o r m a n c e characteristic of i n t e r e s t . The p r o p a g a t i o n of v a r i a n c e
formula i s

(4-10)

where
2
(si is t h e v a r i a n c e o f the performanceparameter Y
i

2
a x is a v a r i a n c e of t h ei n p u tp a r a m e t e r X
j
j
N i s t h e number o fc o n t r i b u t i n gi n p u tp a r a m e t e r s , and
-
X
j
i s t h e mean v a l u eo f X
j
.
The term p i s a c o r r e l a t i o nc o e f f i c i e n tt h a t relates t h ep a r a m e t e rc o n t r i b u -
rs -
tions X
r
and X
SY
and t h es u b s c r i p t s (?r
j
Xr, and ?rS ) indicate the points a t which
thepartialderivativesfortheseinputparameters are o b t a i n e d .
The f i r s t term i n Eq. (4-10) i n c l u d e st h ev a r i a n c eo fe a c hi n p u tp a r a m e t e r and
thepartialderivativeoftheperformancemeasurewithrespecttothatparameter.
S i n c et h ef a c t o r si nt h i s term are s q u a r e dt h e y are a l l p o s i t i v e . The secondterm
intheequationcanbeeitherpositiveornegative; i t i n c l u d e se a c hp a i ro f
c o r r e l a t e dp a r a m e t e r s .T h i s term s i m u i a t e s t h e t r u e s i t u a t i o n i n w h i c hc o r r e l a t i o n
between two i n p u t p a r a m e t e r s c a n e i t h e r i n c r e a s e o r d e c r e a s e t h e t o t a l p e r f o r m a n c e
variability. From t h i se q u a t i o nt h ev a r i a n c eo f any performancemeasure canbe Y
j
obtainedfromknowledge of t h e mean, v a r i a n c e , and c o r r e l a t i o n c o e f f i c i e n t s of e a c h
input parameter.
Inthepropagation-of-variance method a l l o u t p u t v a r i a b l e s a r e assumed t o b e
linearfunctions of t h e i n p u t v a r i a b l e s , and a l l i n p u t p a r a m e t e r d i s t r i b u t i o n s are
assumed tobenormal. Hence,non-normal i n p u tp a r a m e t e rd i s t r i b u t i o n s are approxi-
mated by normalones i nt h ep r o p a g a t i o n - o f - v a r i a n c ef o r m u l a . As seenfrom Eq. (4-lO),
t h e method r e q u i r e s t h e c a l c u l a t i o n o f p a r t i a l d e r i v a t i v e s . T h i s c a n b e done i n pre-
cisely the same way t h a t t h e p a r t i a l d e r i v a t i v e s are c a l c u l a t e d f o r w o r s t - c a s e a n a l y s i s .
Possiblesourcesforvalues of moments o f t h e i n p u t p a r a m e t e r d i s t r i b u t i o n s are
manufacturersdata,testing a l a r g e number ofcomponents, orassumptionsbased on
e x p e r i e n c e .F o re x a m p l e ,r e c o r d i n ga n dp l o t t i n gt h er e s i s t a n c ev a l u e s of a l a r g e
number of r e s i s t o r s o f a given nominal value w i l l produce a p l o t , known as a h i s t o g r a m ,
as shown i n F i g . 4.1. I nt h ef i g u r et h ew i d t h so ft h e small r e c t a n g l e s , c a l l e d cells,
representequalincrementsofresistancevaluesthatfallwithintheindividual
 Value
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
I
1
1
kilohms
F i g u r e 4-1. H i s t o g r a m o f R e s i s t a n c e V a l u e s f o r a Resistor

r e s i s t a n c ei n c r e m e n t s . The sum of t h e h e i g h t s of a l l t h e c e l l s e q u a l s t h e t o t a l
number of r e s i s t o r s t e s t e d . A mean v a l u e f o r t h i s i n p u t p a r a m e t e r , namely r e s i s t a n c e ,
c a nb ec a l c u l a t e d by a d d i n g t o g e t h e r t h e r e s i s t a n c e s o f t h e i n d i v i d u a l u n i t s a n d
d i v i d i n gt h e sum by t h e t o t a l number of u n i t s . The v a r i a n c e , 02, i s c a l c u l a t e d by
t a k i n g t h e number o f r e s i s t o r s i n e a c h c e l l and multiplying each by t h e s q u a r e o f
thedifferencebetweenthemidcellvalueand mean v a l u e ; t h e s e p r o d u c t s are t h e n
a d d e da n dd i v i d e db yt h et o t a l number of r e s i s t o r s t o g i v e a'. The s q u a r er o o to f
thevarianceyo , c a l l e dt h es t a n d a r dd e v i a t i o n , is f r e q u e n t l y u s e d t o d i s c u s s t h e
dispersion of normal frequency distributions.
4.2.4 The Convolution Method
The c o n v o l u t i o n method i s a n o t h e r a t t e m p t t o a c c o u n t f o r t h e s t a t i s t i c a l
d i s t r i b u t i o nc h a r a c t e r i s t i c s . The s i m p l i f i e df o r md i s c u s s e dh e r ea l s o relies on t h e
p a r t i a ld e r i v a t i v e s as computedabove.Thisapproachhasgenerallyfoundonlylimited
practical application; a computerimplementationandcomparisonwithothertechniques
is d e s c r i b e di nR e f . 4-6.
The c o n v o l u t i o n method d e s c r i b e d i n R e f . 4 - 6 is a s p e c i a l i z a t i o n o f t h e more
g e n e r a la n a l y t i c a la p p r o a c hd e s c r i b e di n Ref. 4-7. The b a s i sf o rt h ec o n v o l u t i o n
method is t h e a s s u m p t i o n t h a t t h e t o t a l v a r i a t i o n i n a n o u t p u t p e r f o r m a n c e p a r a m e t e r
is t h e sum of t h ed e v i a t i o n sc a u s e d by e a c hi n p u tp a r a m e t e ri n d e p e n d e n t l y .T h i s is
analogoustotheassumptionthatno mixed p r o d u c t terms o f t h e T a y l o r series are

.L,. ... ...

required. When limited to linear terms only, the partial derivatives represent the
functional relationship between each individual parameter contribution and the para-
meter variations. For a particular interval of the total range of variation of the
output performance parameter, the corresponding interval of each of the model para-
meters can be determined by obtaining the inverse of the partial derivative.
The
probability that the output parameter lies in a particular interval is the sum of
the relative probabilities for the individual input parameters in their. respective
and corresponding intervals. Repeating this process over the appropriate intervals
to cover the total range of variation yields a histogram representation for each
output parameter. Since the convolution method does not assume normal distributions,
it can be used to advantage when input parameters
knownare
to have distributions
differing significantly from normal.
4.2.5 Monte Carlo Analysis
The theory of the Monte Carlo approachPVAtois based on a statistical theorem
called the Glivenko-Cantelli theorem [Ref.
4-21 which is:
Given a function of n random variables,
Y = f(X1, X 2 , ..., Xn) with each
variable X. described by a distribution, then select a value forX each
i'
i=l, 2, ...,n, from their respective distributions and compute a value of
Y. Repeat this procedure for m times.As m tend's to infinity, the distri-
bution of Y obtained approaches the actual distributionY. of
In contrast to worst-case analysis which obtains only end-limit values and to
propagation-of-variance analysis whi.ch assunes normal distributions only, a Monte
Carlo analysis determines the actual statistical distributions
of the cutput variables.
The Monte Carlo method permits computer simulation
of a brute-force empirical approach.
The empirical. approach would require the actual construction from representative
components of many copies of the system under study.
A s many copies would be made
and operated as required to obtain good statistical estimates of the system output
variables and the variations in these variables. This empirical approach is usually
highly impractical, and it is seldom if ever applied.
By using a digital computer to simulate the above empirical technique, many of
the objectionable features are removed. Given the mathematical model of a system
under study and a description of the component part populations, it is possible by
doing enough simulations to obtain
to any reasonable degree of accuracy the distri-
butions of the performance measures.
The Monte Carlo method requires the complete
statistical distributions of the input variables at some particular
t. time
The
computer randomly selects a value for each input parameter from its distribution
and uses this value in computing the solution.
The values from a multi-parameter
part cannot be given simply as distributions. Instead they must be in
listed
a way

37
such that a l l theparameterreadingsfromthe same p a r t are g r o u p e d t o g e t h e r w i t h any
necessarycorrelations. Then t h e Monte C a r l o methodmakes a s i n g l e random s e l e c t i o n
from t h i s l i s t i n g which determines a l l the correlated parameter values for the multi-
parameter part.
I n theGlivenko-Cantellitheorem,eachofthe random v a r i a b l e s Xi, which are i n
t h i s case t h e s y s t e m m o d e l i n p u t v a r i a b l e s , c a n h a v e e i t h e r a continuousprobability
d e n s i t yf u n c t i o no r a d i s c r e t ep r o b a b i l i t yd e n s i t yf u n c t i o n . B e c a u s eo n l yd i s c r e t e
quantitiescanbeusedincomputer,onlydiscreteprobabilitydensityfunctions are
of i n t e r e s t t o t h e Monte C a r l o methodof analysis. A d i s c r e t ep r o b a b i l i t yd e n s i t y
f u n c t i o n is simply a normalizedhistogram. Shown i n F i g . 4.2 i s thenormalized
v e r s i o n o'f thehistogramofFig.4.1.

-
30
300

-
20
300

t I

18 100 102 104 X=Resistance,Kilohrns

F i g u r e 4-2. Normalized Form of Fig. 4-1; P r o b a b i l i t y

DensityFunctionfor Discrete Random V a r i a b l e X

Given a d i s c r e t e d i s t r i b u t i o n o fa s s o c i a t e d ( f ( X ) , X) v a l u e s a s shown i n Fig.4.2,

the discrete random v a r i a b l e X p o s s e s s e s t h e p r o p e r t i e s :
Given t h e v a l u e s X = X and X = Xb,
a
p[Xa 5 X 5 Xb ] is t h e p r o b a b i l i t y t h a t
X L X and X 5 X , , where Xa s Xb.
a
Also,

38
where F(X ) is a point on the discrete cumulative distribution. The summation applies
n
to those values of the random variable X which are less thanto the X
or equal
n
specified in the summation. T.he cumulative distribution for the random variable
4..3.
X of Fig. 4.2 is shown in Fig.

0.75

0.5

0.25

Figure 4-3. Cumulative Distribution for Discrete

Random Variable of
X Fig. 4-2

An important pointto understanding the Monte Carlo method is the observation

that an area under the probability density curve amounts to on
a the
point
cumulative
distribution curve. Thus, given a density function
f(X) for which the total area
0.2, then the valueF(X) at X=Xn is 0.2. In a Monte Carlo
between X=O and X=X is
n
computer program the computer converts all the probability density functions to
cumulative distributions. Then the computer generates random numbers and associates
each of these numbers with a particular point on each cumulative distribution.
The
random numbers in this context are numbers chosen at random in the range between
0 and 1.
In order to obtain reasonable accuracy with the Monte Carlo technique a large numbe
of randomly-generated replicas of the system are made;for the solution to each replica the

39
o u t p u tp a r a m e t e r so fi n t e r e s t are obtainedfromthesystemmodelequations. The t o t a l
number o f s y s t e m s o l u t i o n s ( a l s o c a l l e d s y s t e m s i m u l a t i o n s ) r e q u i r e d is o b t a i n e d v i a
a tradeoffbetweenaccuracyandthecostofcomputer time. T h i s number c a nv a r y
anywherefrom 50 t o 5,000 o r more d e p e n d i n g o n t h e p a r t i c u l a r a p p l i c a t i o n . The
number o fs o l u t i o n st y p i c a l l yu s e df o ro n ep r o g r a m is 500 [Ref. 4-21. When p r a c t i c a l ,
a profession statistician should be consulted on how t o a r r i v e a t a n a p p r o p r i a t e
number of s i m u l a t i o n s f o r a givensystemandpurpose.
Once a l l t h e Monte C a r l o s o l u t i o n s h a v e b e e n g e n e r a t e d , t h e p r o b a b i l i t y d e n s i t y
f u n c t i o n sf o re a c ho ft h eo u t p u tp e r f o r m a n c em e a s u r e sc a nb eo b t a i n e d .S i n c et h e
completedistributionforeachoutputvariable is available, coefficients whichde-
scribethevarious s t a t i s t i c a l p r o p e r t i e s of t h e d i s t r i b u t i o n s c a n b e computed as
required.
I t s h o u l db er e c o g n i z e dt h a td u et ot h el a r g e number o fs y s t e ms i m u l a t i o n s re-
quired,the Monte C a r l o t e c h n i q u e is best suited to variation analysis ofsystems
which cannotbehandled by less b r u t e - f o r c et e c h n i q u e s . Its c o s t and time l i m i t a t i o n s
must be c o n s i d e r e db e f o r ec h o o s i n gt h et e c h n i q u ef o r a p a r t i c u l a rs y s t e m .I n a circuit
a n a l y s i sp r o g r a m ,f o re x a m p l e ,d cs o l u t i o n sc a nb eo b t a i n e d a t reasonablecostvia
t h e Monte C a r l o t e c h n i q u e ; Monte C a r l o ac s o l u t i o n s are u s u a l l y l e s s p r a c t i c a l .
Finally,therelativelylarge amount ofcomputer time r e q u i r e d f o r a singletransient
solutionof a c i r c u i t means t h a t i t is u n r e a s o n a b l e t o a t t e m p t t o o b t a i n s u f f i c i e n t
transient solutions to make t h e Monte C a r l o t e c h n i q u e a practicalapproachtoobtaining
d i s t r i b u t i o n s of circuitperformancemeasuresrelatedtotransientresponses.
An i n t e r e s t i n g v a r i a t i o n on t h e Monte Carlotechniquehasbeenreported[Ref. 4-21.
It combinesportionsofworst-case and Monte C a r l oa n a l y s i s .O f t e nt h ed a t ag i v i n g
t h ea c t u a ld i s t r i b u t i o n so fi n p u tp a L x e t e r s are n o ta v a i l a b l e . What hasbeendone
inthecitedreferencefor,suchcases is to substitute a rectangulardistribution
whose upperandlower limits are theupperandlowerworst-case limits. A Monte
Carloanalysis i s thenperformed,whichprovides a b e t t e r estimate of c i r c u i t p e r f o r -
mance t h a n would b eo b t a i n e d by u s i n gt h ec o n v e n t i o n a lw o r s t - c a s ea n a l y s i s .S i n c e
theactualinputparameterdistributions are n o t r e c t a n g u l a r , t h e p r o b a b i l i t y of
selectingvaluesclosetotheworst-casevalues is g r e a t e r t h a n f o r t h e a c t u a l d i s t r i -
b u t i o n s .C o n s e q u e n t l y ,t h er e s u l t i n gd i s t r i b u t i o n s a r e less o p t i m i s t i ct h a n would
beobtainedfromtheactualinputdistributionsbut are n o t a s p e s s i m i s t i c as worst-
case solutions.
4.3 PVA ComputerPrograms
Many computerprograms exist for i m p l e m e n t i n gi n d i v i d u a l l yt h e PVA t e c h n i q u e s
d i s c u s s e di nS e c . 4.2. Some oftheseprograms are l i s t e d i n T a b l e 4-1. However,
r e l a t i v e l y few a r e known t o e x i s t which are a v a i l a b l e o u t s i d e t h e o r g a n i z a t i o n s

40
 Table 4-1
Programs i n t h e PVA Area

Organizations(0riginator
Program Code Userlsponsor)
Program or
Description References
PV-RTI -PerformanceVariationanalyses;generalprogram RTIINASA 4-3
f o rw o r s t - c a s e , moments, simulation, e t c .
MCS-IBM -
Monte C a r l oS i m u l a t i o nf o rp e r f o r m a n c ev a r i a t i o n IBM/AF-RADC 4-8
analysis with programmed f u n c t i o n a l model
MCS-GDC -
Monte C a r l oS i m u l a t i o nf o rp e r f o r m a n c ev a r i a t i o n GD-Convairl? 4-9
a n a l y s i s w i t h programmed f u n c t i o n a l model
PV-LS -
Performance Variation analysis program for systems Lear SieglerINASA 4-10
PV-SE -Performance Variation analysis program using Monte SylvaniaElectronics/ 4-11
Carlosimulationwith programmed mathematical modelAF-WC
A
CI MANDEX-NAA -
Modified AND Expanded worst-case method f o r a n a l y s i s MI? 4-2
of c i r c u i t performance v a r i a t i o n s w i t h c i r c u i t
equations
"-NAA -
Moment Method f o rc i r c u i p
t e r f o r m a n c ev a r i a t i o n -
NAAI? 4-2
analysiswithcircuitequations; computer mean and
v a r i a n c e ;c o r r e l a t i o ni n c l u d e d
MCS-NAA -
Monte C a r l o
gimulation
f ocr i r c u ipt e r f o r m a n c e -
NU/? 4-2
variationanalysiswithcircuitequations;corre-
lation included
VINIL-NAA
%N +
I method f o rc i r c u i t performancevariation -
NU/? 4-2
analysiswithcircuitequations
PW-NAA -
Parameter Variation Method f o r c i r c u i t p e r f o r m a n c e N&/? 4-2
variationanalysiswithcircuitequations; one-at-
a-time andtwo-at-a-time analyses
where they originated and which combine several PVA t e c h n i q u e s i n t o a s i n g l e program.
A FORTRAN l i s t i n g o f a g e n e r a l PVA program which implements nearly a l l of the pvA
techniquesdiscussedinSec. 4.2 is g i v e n in Appendix A; i t i s d e s c r i b e d i n some
d e t a i l below.
Two w i d e l y u s e d c i r c u i t a n a l y s i s p r o g r a m s w h i c h h a v e some PVA c a p a b i l i t i e s a r e
ECAP and NASAP. The E l e c t r o n i cC i r c u i tA n a l y s i s Program (ECAP) i s a v a i l a b l e t o
users of IBM computers. TheNetwork A n a l y s i sf o rS y s t e mA p p l i c a t i o nP r o g r a m (NASAP)
is a NASA program. Althoughworking a t a number ofcomputer i n s t a l l a t i o n s , NASAP
is s t i l l i n development.These two programsand t h e i r PVA c a p a b i l i t i e s are d i s c u s s e d
later i n t h i s s e c t i o n .
4.3.1 A G e n e r a l PVA Program
A flowdiagramof a g e n e r a l PVA program i s shown i n F i g . 4-4. A s c a nb es e e n
from t h e f i g u r e , t h e program i s keyed t o t h e s u b r o u t i n e w h i c h e v a l u a t e s t h e p e r f o r m a n c e
model. To make t h ep r o g r a ma p p l i c a b l et o any kindofsystem,nobuilt-inperformance
m o d e ls u b r o u t i n e is i n c l u d e d ; t h i s s u b r o u t i n e mustbesupplied by t h e u s e r of t h e
program[Ref. 4-31.
The i n p u t t o t h e program is a m a t h e m a t i c a l d e s c r i p t i o n of t h e s y s t e m model
(and t h e time b e h a v i o r o f t h e m o d e l , i f r e q u i r e d ) , t h e number of random v a r i a b l e s
and t h e number of f i x e d v a r i a b l e s i n v o l v e d , t h e means o r n o m i n a l v a l u e s of t h e i n p u t
variables,thestandarddeviationsorstepsizesintheinputvariables,theinput
variabledistributions,ifavailable, and t h e c o r r e l a t i o n s o f t h e i n p u t v a r i a b l e s .
An a d d i t i o n a l i n p u t t h a t is requiredfor some a n a l y s e s i s a s e l e c t i o n o f v a l u e s o f
t h ee l e m e n tp a r a m e t e r s a t whichtheperformancemodel is t o b e e v a l u a t e d . I f t h e s e
v a l u e s are s e l e c t e d m e t h o d i c a l l y a c c o r d i n g t o some s t a t i s t i c a l d e s i g n , t h i s a l l o w s
or efficient generation of t h e o u t p u t s t o u s e i n a multipleregressionanalysis.
Monte C a r l o S i m u l a t i o n
A Monte C a r l o s i m u l a t i o n i s used t o estimate t h e p e r f o r m a n c e d i s t r i b u t i o n i n
terms o f t h e i n p u t d i s t r i b u t i o n s , c h a r a c t e r i s t i c s , etc. I ft h ei n p u tv a r i a b l e s are
normally distributed the means,standarddeviations,andthecorrelationmatrix are
r e q u i r e d .I ft h ei n p u tv a r i a b l e s are n o tn o r m a l l yd i s t r i b u t e dt h ea p p r o p r i a t ed i s -
t r i b u t i o n c h a r a c t e r i s t i c s m u s tb es p e c i f i e d . The p r o g r a mh a sp r o v i s i o n sf o rh a n d l i n g
any one of t h e f o l l o w i n g d i s t r i b u t i o n s :
Uniform,
(1)
(2) Normal,
(3) Log-Normal ,
(4) Exponential,
(5) Weibull ,

42
 INPUT
Number of Variables X
-- - - """ - - - - -i- -" "

Random Variables
Parameters
. Means . Nominal Values
. StandardDeviations . S t e pS i z e s
Generate . Distribution Form
- G e n e r aSt et a t i s t i c a l
Random Uniform
Design
Variable Time Behavior

Random Variables Subroutine

with for Evaluating 4 Inputs Fixed Generate
Appropriate Distribution Y = G(X,t)
b
1

IJI
Interaction Analysis I

1
J(
S e n s i t i v i t y , Worst-case Least-Squares Analysis
of D i s t r i b u t i o n and Moment Analysis of Computed Performance Attr-ibutes
~~~ ~ ~~ ~~

-Mean -Calculate Parial Derivatives Sensitivity

-Variance -Taylor S e r i e s Approximation S i g n i f i c a n tI n t e r a c t i o n s
-StandardDeviation -Worst-case Analysis Worst-case Analysis
-Third and Fourth Moments -Sensitivity
-Skewness -Checks for Non-linearity
-Kurtosis -Standard Deviation of
-Covariance Matrix Performance A t t r i b u t e s
-Ranking
V
i t Appropraite Distribution
Edgeworth series
LaguerrePolynomials

Figure 4 - 4 . Flow Diagram forGeneral PVA Program

 ,
(6) Gamma (Integral values of one parameter)
(7) Chi-square ,
(8) Triangular, and
(9) Beta (Integral values of both parameters).
Uniformly distributed variables are first generated; they are then transform
according to the methods described in Ref. 4-3, Appendix B to variables having t
appropriate distributions as specified in the input. These transformed variables
are then used to compute the performance measures such as voltage output, curre
output, power dissipation,etc. The performance measures are generated the number
of times required to obtain 'the desired precision of the results. When the inputs
are preciselyknown, the number of trails necessary to estimate the distributio
function of a performance measure to the required degree of precision for a one
dimensional distribution can
be estimated from the Kolmogoroff-Smirnov statistic
d between the sampled distribution function and the true
for the maximum deviation
(but unknown) distribution function. Table 4-2 displays the number of observations
necessary in order that the probability be the maximum deviatian between the
a that
distribution function and the sample function exceeds thed. value

Table 4-2
Percentiles of the Distribution d of
for Several Values of 1-a
-
-
1-CC
N 0.80
0.99
0.95
0.90
0.85

0.45 5 0.47 0.51 0.56 0.67

10 0.32 0.34 0.37 0.41 0.49
0.23 20 0.25 0.26 0.29 0.35
30 0.19 0.20 0.22 0.24 0.29
40 0.17 0.18 0.19 0.21 0.25
50 0.15 0.16 0.17 0.19 0.23

For larger values of N 1.071.14-

1.22 1.36. .1.63
fi hi fi hi fi

44
 Hence, i f N is 50 t h e c h a n c e i s 0.05 t h a t t h e m a x i m u m d e v i a t i o n b e t w e e n t h e s a m p l e
distribution function and t h e a c t u a l d i s t r i b u t i o n f u n c t i o n e x c e e d s 0.19; i f N = 100,
d = 0.136,and i f N = 1000, d = 0.043. When h i g h p r e c i s i o n i s needed, i t is p o s s i b l e
t o perform a v e r y l a r g e number o f s i m u l a t i o n trials. However, i t mustbe remembered
that the cost i n computer time p e r s i m u l a t i o n d e p e n d s on t h e c o m p l e x i t y o f t h e p e r -
formance model subroutine.
In practice the distributions of the component c h a r a c t e r i s t i c s a r e s e l d o m known
v e r yp r e c i s e l y . Hence t h e r e i s a p r e c i s i o n of t h e d i s t r i b u t i o n o f t h e p e r f o r m a n c e
measurebeyondwhich i t is i m p r a c t i c a l . t o a t t e m p t o r e f i n e t h e estimate of t h e t r u e
distribution. I n fact, v e r yo f t e n a u n i f o r md i s t r i b u t i o no ft h ei n p u tv a r i a b l e is
assumebecauseofthelackofknowledgeconcerningthetruedistribution.
Suppose now t h a t a r a t i o n a l p r o c e d u r e is available for estimating N and t h a t N
valuesoftheperformanceshavebeencomputed. Then t h e N o b s e r v a t i o n sa r er a n k e d
inascendingorderofperformance,theirfirstfourcentral moments a r e computed,
and t h em e a s u r e so fs k e w n e s sa n dk u r t o s i sa r eo b t a i n e d . From t h e s t a t i s t i c s i t can
bedecidedwhichdistributiontofit to t h e d a t a o r which series a p p r o x i m a t i o n s t o
use. The a p p r o x i m a t i n g d i s t r i b u t i o n s c a n b e f i t t e d by t h e methodof moments.
I n t h i s program t h e Edgeworth series and/or Laguerre polymonials are used t o
a p p r o x i m a t et h e unknown d i s t r i b u t i o n f u n c t i o n . Themethods forfittingthesedis-
t r i b u t i o n s are givenbyKendall[Ref. 4-12].
S e n s i t i v i t y and Moment A n a l y s i s
ThisprogramobtainsTaylor series a p p r o x i m a t i o n t o t h e m o d e l s and a s i l l u s t r a t e d
i n Fig. 4-4 s u b s e q u e n t l y u s e s them t o p r e d i c t w o r s t - c a s e p e r f o r m a n c e s , t o e s t i m a t e
s e n s i t i v i t i e s ofperformancemeasures toinputs,to check f o r n o n l i n e a r i t i e s and
i n t e r a c t i o n s ofbehaviorwith respect t o i n p u t s , and t o p e r f o r m a moment a n a l y s i s .
The s t e p s i z e s are chosen t o i n c l u d e t h e e x p e c t e d r a n g e o f v a r i a t i o n o f t h e i n p u t
v a r i a b l e s as a r e s u l t of t h e e n v i r o n m e n t s d e s c r i b e d by t h e m i s s i o n p r o f i l e , t h e
inherentvariationsinthepartcharacteristics, and t h e a g i n g e f f e c t s .
Thispartoftheprogramfirst computes e s t i m a t e s of t h e f i r s t andsecond partial
d e r i v a t i v e s of t h e p e r f o r m a n c e m e a s u r e s o f i n t e r e s t w i t h r e s p e c t t o e a c h o f t h e
p e r t i n e n tp a r tc h a r a c t e r i s t i c s ,i n p u t s ,l o a d s , e t c . ; t h ef i v e - p o i n tc e n t r a ld i f f e r e n c e
formulasareusedforobtainingthepartialderivatives.
Having o b t a i n e d t h e f i r s t andsecond p a r t i a l d e r i v a t i v e s ofa performance measure
w i t hr e s p e c tt ot h ei n d e p e n d e n tv a r i a b l e s ,t h ef o l l o w i n gT a y l o r series expansion i s
obtained .
Y(hl,hp, * -- 9 hm ) = YN + CYi1
I
1
h. -+ 7 CYi
I' 2
hi + ... (4-11)

45

I. -
"

where
YN is t h en o m i n a lv a l u e of performancemeasure Y,

Y i & Y y are r e s p e c t i v e l y t h e 1-st and 2-nd p a r t i a l s of Y w i t h r e s p e c t t o i n p u t

v a r i a b l e Xi,
hi is thechangefromnominal of Xi, a n dt h e sums are o v e r a l l m i n p u t v a r i a b l e s .

D i v i d i n g by YN y i e l d s

- = 1+
V

CLSi + CQS, , (4-12)

n'

where
LSi = a measure of l i n e a r s e n s i t i v i t y o f t h e p e r f o r m a n c e m e a s u r e t o t h e
i-th input variable
1
Yi hi
LSi = - (4-13)
yN
and
QSi = a measure of second d e g r e e o r q u a d r a t i c s e n s i t i v i t y ( d e n o t e d as n o n l i n e a r
sensitivity in the programoutput) of t h e p e r f o r m a n c e w i t h r e s p e c t t o t h e i - t h i n p u t
v a r i a b l e and is given by

2
QSi = T1 Yi"
hi /YN. (4-14)

These two q u a n t i t i e s are p r i n t e d o u t f o r e a c h o f t h e N variables. The s e n s i t i v i t y

measureassociatedwiththei-thvariable is e s s e n t i a l l y t h e r e l a t i v e change i n
theperformancemeasure as a f u n c t i o n o f t h e maximum e x p e c t e d c h a n g e i n t h e i - t h
variable. The d e f i n i t i o n s o f s e n s i t i v i t y and n o n - l i n e a r i t y are s u g g e s t e d by t h e
T a y l o r series expansion. As n o t e d e a r l i e r , t h e r e are s e v e r a l - d e f i n i t i o n s of
s e n s i t i v i t ya p p e a r i n gi nt h el i t e r a t u r e . The d e f i n i t i o n s u s e d i n t h i s program are
very convenient in estimating the relative change in a performancemeasure Y for
theexpectedchangesintheindependentvariables.
The T a y l o r series expansion as p r e s e n t e d a b o v e d o e s n o t i n c l u d e terms w i t h
mixed p a r t i a l d e r i v a t i v e s . To o b t a i n t h e s e c o n d p a r t i a l d e r i v a t i v e s w i t h r e s p e c t
t o a l l p a i r s of i n d e p e n d e n t v a r i a b l e s would r e q u i r e c o n s i d e r a b l y morecomputing
time. The computation i s p e r f o r m e du s i n go n l yt h ef i r s tp a r t i a l s and t h ep u r e
s e c o n dp a r t i a l s ;t h e series approximation is t h e nc h e c k e df o r i t s adequacy. I ft h e
r e s u l t s are n o t as p r e c i s e as r e q u i r e d , t h e a p p r o p r i a t e mixedsecond p a r t i a l s are
o b t a i n e d by a program described i n t h e s e c t i o n on I n t e r a c t i o n A n a l y s i s .

46
Worst-case Limits
The worst-case limits are computedby the procedure described by West and S c h e f f l e r
rRef.4-131. The s i g n s o f t h e f i r s t p a r t i a l d e r i v a t i v e s are examined; t h e v a r i a b l e s
for which they are p o s i t i v e are p l a c e d a t t h e i r h i g h v a l u e s , X + h , and t h e v a r i a b l e s
f o r whichthey are n e g a t i v e , a t t h e i r low v a l u e s , X - h , inordertoestimate
an upperworst-case limit. Conversely, t o e s t i m a t e a lower l i m i t t h e v a r i a b l e s f o r
which Y ' is p o s i t i v e are p l a c e d a t t h e i r low v a l u e s , and f o r Y' n e g a t i v e , a t t h e i r
h i g hv a l u e s . Theworst-case limits oftheperformancemeasures are computedby
actually substituting the appropriate values of t h e v a r i a b l e s i n t o t h e f u n c t i o n s
c o m p r i s i n gt h ep e r f o r m a n c e model.Thecomputedworst-case limits are t h e n compared
tothe limits e s t i m a t e dw i t ht h eT a y l o r series e x p a n s i o n .I ft h e s ev a l u e s do n o t
agreetowithintherequiredaccuracy,theomitted terms, namely, t h e mixed p a r t i a l
d e r i v a t i v e s( i n t e r a c t i o n s ) and t h eh i g h e ro r d e rp u r e t e n u s mustbe investigated. The
higherorderpurederivatives.areconvenientlycheckedonevariable a t a time by
comparing t h e f u n c t i o n a l v a l u e a t t h e two end p o i n t s w i t h t h a t e s t i m a t e d b y t h e
f i r s t a n ds e c o n dp a r t i a l sw i t h respect t ot h a tv a r i a b l e .T h e s ec h e c k ss u g g e s tt h e
source of any l a c k o f p r e c i s i o n .
Moment A n a l y s i s
The moments of the performance measures can be obtained from the Monte C a r l o
simulationrunsorfrom an e r r o r p r o p a g a t i o n a n a l y s i s b a s e d on t h e T a y l o r series
approximation. The l a t t e r is s i m p l e r t o computeand n o ts u b j e c tt os a m p l i n gf l u c t u a -
t i o n s as is t h ef o r m e r . However, t h e series approximation i s s u b j e c t t o t h e l a c k
of p r e c i s i o n w i t h which i t a p p r o x i m a t e s t h e t r u e f u n c t i o n .
Let

If only the first order terms a r e u s e d , t h e e s t i m a t e s of t h e mean and v a r i a n c e

2
of Y , denoted by F{Y) and 8 {Y) r e s p e c t i v e l y , are givenby

1% ay

""j

47
where
A

SIX,) = estimated standard deviationof the measurements X

iy

p IXi,Xj) = X
estimated simple correlationof the measurements on X and
i j

If Xi and X are characteristics of two distinct components, thenp {X. ,X.}= 0 ;

j 1 J
otherwise, it isestimated by

If the first and second order terms (not including the mixed partials-interaction
terms) are used in the approximation, then further terms are required in the momen
analysis.
Let

then the estimated mean and variance

for Y can be written as

+ ICY Y
i j
cov~xi,x.l
J

1 1 It
+ $ CYi Y
i

48
 where E{X) d e n o t e s t h e e x p e c t e d o r mean v a l u e o f X and $3i and i?4i are t h e e s t i m a t e d
t h i r d and f o u r t h moments of Xi, i=l, ...,m. A similar expansion may b e o b t a i n e d w i t h
theinteractiontermsincluded.
In the above a n a l y s i s i t h a s i m p l i c i t l y b e e n assumed t h a t t h e r e l a t i o n s h i p
between the performance measure Y and t h e p a r t c h a r a c t e r i s t i c s , Xi, ..,m
i=l,. is
known, t h a t i s , t h ec o e f f i c i e n t s are known. However, i n p r a c t i c e t h e r e l a t i o n s h i p
may b e o b t a i n e d f r o m e m p i r i c a l d a t a a n d t h e c o e f f i c i e n t s may b e c o n s i d e r e d estimates
of t r u e b u t unknown v a l u e s . The e x t e n t t o which t h e d a t a are a v a i l a b l es h o u l dt h e n
b er e f l e c t e di nt h ep r e c i s i o n so ft h ei n p u t st ot h e error p r o p a g a t i o na n a l y s i s . A
completediscussionofthisproblem is g i v e n i n M a r i n i , Brown, and Williams [Ref.4-14].
Interaction Analysis
Incasetheworst-case limits computed d i r e c t l y from t h e f u n c t i o n s are notade-
quatelyapproximated by t h e l i n e a r and p u r e q u a d r a t i c terms, i t i s n e c e s s a r y t o
compute t h e mixed p a r t i a l d e r i v a t i v e s f o r t h e p a i r s o f v a r i a b l e s which are expected
toyieldsignificantinteractioneffects. Themixed p a r t i a l sc a nb e computedby
one of t h e f o l l o w i n g two methods.
One procedure would b e t o compute t h e f i r s t p a r t i a l d e r i v a t i v e s w i t h r e s p e c t
tothei-thvariable a t f i v ed i f f e r e n tv a l u e so ft h ej - t hu a r i a b l e .T h e s ep a r t i a l s
would i n t u r n b e u s e d t o compute t h es e c o n dp a r t i a l .T h i sp r o c e d u r ea s s u m e s a
d e g r e eo fs m o o t h n e s so ft h ea n a l y t i c a lf u n c t i o n .
A secondprocedurewouldbe to generate the performance measure for selected
sets o f v a l u e s o f t h e i n d e p e n d e n t v a r i a b l e s a n d t h e n f i t by r e g r e s s i o n t e c h n i q u e s
t h ef u n c t i o n a l form

Y = b0 + C biXi + Z biiXf + C C bijXiXj .

Thisassumes a l l h i g h e ro r d e re f f e c t sc a nb ea d e q u a t e l ya c c o u n t e df o r by a second
degreepolynomialfunction. The c o e f f i c i e n t s of t h e terms X.X. would correspond
= J
tothe mixed p a r t i a l su n d e rt h ea s s u m p t i o n . The s e l e c t i o n of t h ev a l u e so ft h e
variablescanbeperformedefficiently by t h e methodof statistical designs for
f a c t o r i a le x p e r i m e n t s . Methods f o rg e n e r a t i n gt h ea p p r o p r i a t ed e s i g n are d e s c r i b e d
by Addelman [Ref.4-15].
4.3.1.1General PVA Program Example
The g e n e r a l PVA programwhichhasbeendiscussed in the preceding pages
canbeused t o perform a w i d e v a r i e t y o f v a r i a t i o n a n a l y s e s f o r a wide variety of
systems. Some examplesusingthisprogram are given i n Ref. 4-3. A simpleexample
is r e p r o d u c e d h e r e f r o m t h a t r e f e r e n c e .
A second degree polynomial was chosen for illustration of t h e program.

49

1.
 Y = 1 +2x1 + 2x2 + 3x1x2 + 4x12 + 4x22 .
T h e r e are two i n d e p e n d e n t v a r i a b l e s , X1 and X2, and one dependent variable Y denoted
by POLY i n t h e programinput. One hundred(100)simulation trials were performed
assuming X1 and X are n o r m a l l y d i s t r i b u t e d w i t h means 10 and 5 a n d s t a n d a r d d e r i v a t i o n s
2
0.2 and 0.05, r e s p e c t i v e l y , and c o r r e l a t i o n 0.5.
1

In the interaction analysis part of t h e p r o g r a m , o n e n e e d s t o i n d i c a t e w h i c h

i n d e p e n d e n tv a r i a b l e s ,f r o mt h o s ea v a i l a b l e , are t o b e u s e d i n t h e a n a l y s i s . I n
t h e s p e c i f i c examplethere are o n l y two s u c h v a r i a b l e s andbothof them are used a s
i n d i c a t e d by i n p u t s 4 and 5.. Ifthere were 1 0 v a r i a b l e s i n a l l and o n l y f i v e v a r i a b l e s
t ob eu s e di nt h ea n a l y s i s , e. g.. v a r i a b l e s numbered 1, 3 , 5 , 8, and10,theninput
5 would b e t h e s e n u m b e r s i n t h e a p p r o p r i a t e f o r m a t a n d i n p u t 4 wouldbe NVT = 5 and
NW = 5 provided a l l Z5 combinationsofthe 5 v a r i a b l e s were used.See Addelman
[Ref.4-15] f o r methodsof s t a t i s t i c a l d e s i g n of e x p e r i m e n t s f o r u s i n g a f r a c t i o n of
2
5 runs. The i n p u t sa n do u t p u t sf o rv a r i o u sp a r t so ft h e program are l i s t e d on t h e
f o l l o w i n gp a g e s . The programoutputs are fromthe Bunker-Ram0 340computer; the
program is w r i t t e n i n FORTRAN I1 language.Forconvenience of r e p r o d u c t i o no ft h i s
r e p o r t ,t h ep r i n t o u tf r o mt h ep r o g r a mh a sb e e nr e p r o d u c e d by t y p i n g . The p r i n t o u t
format has been preserved.
ProgramInputDescriptionforSimulation
The f i r s t c a r d h a s t h e s t a r t i n g v a l u e , X
N, forthe random number g e n e r a t o r .
Format (F10.0) .
I n p u tc a r d 2 givesthe numberof models(not more t h a n f i v e ) f o l l o w e d by a
f o u r l e t t e r i d e n t i f i e rf o re a c hm o d e l . Format(I2,5A4).
Thiscardprovidestheactual number of v a r i a b l e s and t h e numberof corre-
l a t e dv a r i a b l e sf o re a c hm o d e l ,a n dt h e number o f s i m u l a t i o n trials f o r
a l l models.
Format
(1115).
T h e s ec a r d sc o n t a i ni n f o r m a t i o nn e c e s s a r yf o r a r e a d a b l eo u t p u t . The f i r s t
c o n t a i n st h e names o f t h e d i s t r i b u t i o n s o f t h e random number g e n e r a t o r s( e a c h
name is l i m i t e dt ot w e l v ec h a r a c t e r s ) . The s e c o n dh a st h e names o ft h e two
polynomial f i t routines,namelyEdgeworthandLaguerre. Format(20A4).
The v a r i a b l e i n p u t c a r d s c o n t a i n n o m i n a l a n d d e v i a t i o n v a l u e s , a parameter
name, and a randomnumber generatorcallvalue. The c a l l v a l u e i s t h e argument
f o r a COMPUTED GO TO s t a t e m e n t and c a l l s t h e a p p r o p r i a t e g e n e r a t o r s u b r o u t i n e .
Format(2E10.4,A4,14).Those v a r i a b l e sw h i c hh a v en o n - z e r oc o r r e l a t i o n sw i t h
other variables mustberead in first.
If there are c o r r e l a t e d v a r i a b l e s , t h e v a l u e s are r e a d as a n u p p e r t r i a n g u l a r
matrix.
Format
(16F5.0).

50
Input Descri-tion for Sensitivity,Worst-Case,
" and Moment Analysis
(1) Model i d e n t i f i c a t i o n is on t h e f i r s t c a r d . Thenumber o fm o d e l s ,n o tt o
exceed 10, i s followed by f o u r l e t t e r model d e s c r i p t o r s . Format(I2,10A4).
(2)The n e x tc a r dg i v e st h ev a r i a b l ei n f o r m a t i o nf o re a c h model.Thenumber
o fv a r i a b l e sf o re a c hm o d e l ,n o tt oe x c e e d2 0 , i s i n Format(1012).
(3614)These c a r d s a r e i d e n t i c a l t o t h e s i m u l a t i o n i n p u t c a r d t y p e s (5) and ( 6 ) .
Thenominalanddeviationvalues(one-halftheexpectedextremedeviation
v a l u e s ) are i n t h e same f o r m a t a n d t h e v a r i a b l e name s h o u l d a l s o b e g i v e n ,
(2E10.4 ,A4) . Correlatedvariables, Format(16F5.0).

Input Description for Interaction Analysis

Cardone is f o r t h e number ofmodels,Format(12).
Card two s p e c i f i e s t h e t o t a l number o fi n d e p e n d e n tv a r i a b l e s (NV) and t h e
(alphanumeric) name f o rt h ed e p e n d e n tv a r i a b l e . Format(I2,A4).
The v a r i a b l e c a r d s s p e c i f y t h e n o m i n a l v a l u e s and d e v i a t i o n s ofeach
i n d e p e n d e n tv a r i a b l e ,a s w e l l as i t s (alphanumeric) name. There is one
c a r df o re a c hv a r i a b l e . Format(2E10.4,A4).
T h i sc o n t r o lc a r di n d i c a t e st h e numberof v a r i a b l e s (NVT) t o b e u s e d i n t h e
interactionanalysis (NVT 5 NV) and t h e numberof v a r i a b l e s whose l e v e l s
a r et ob e computed(NW). I f NVT = NW, a l l combinations are c o n s i d e r e d ;
o t h e r w i s e NW < NVT. Format(212).
Card f i v e i n d i c a t e s , by s u b s c r i p t s , t h e v a r i a b l e s s e l e c t e d f o r a n a l y s i s .
Thenumber of v a l u e s a p p e a r i n g s h o u l d b e NVT i n format(2012).
Card 6 is o m i t t e d i f NVT = NW. Otherwise i t s p e c i f i e s , by s u b s c r i p t s ,
t h e NW v a r i a b l e st ob e computed.Format(2012).
Cards 2-6 a r er e p e a t e df o re a c hm o d e l . The d e v i a t i o n s s p e c i f i e d onCard 3 a r e doubled
f o rt h el e a s ts q u a r e sa n a l y s i s .T h a t i s , t h eu p p e r andlower limits c o n s i d e r e df o r
eachvariablearethe nominalvaluesplusandminus twice t h e d e v i a t i o n s g i v e n on
Card 3 .
Theprogram inputs to the example using the polynomial introduced as t h e p e r f o r -
mance model a r e g i v e n i n T a b l e 4-2 andfollowed by t h e o u t p u t s i n T a b l e s 4-3 through
4-5.

51
 T a b l e 4-2
Program I n p u t sf o rP o l y n o m i a l (POLY) Example

I n p u t s (CardImage)

Simulation Analysis
(1) 1697.
(2) lP0LY
(3 1 2 2 100
(4) UNIFORM NORMAL LOG NORMAL EXPONENTIAL
WEIBULL GAMMA BETA
CHI SQUARE
EDGEWORTH LAGUEEXE
(5) .1000E 02 .2000E 00 x1 2
.5000E 01 .5000E-01 x2 2
(6) 1 . 0 0.5 1.0

S e n s i t i v i t y ,W o r s t - c a s ea n d Moment A n a l y s i s
(1) lP0LY
(2) 2
(3 1 .1000E 02 .2000E 00 X1 2
.5000E 01 .5000E-01 X2 2
(4) 1.0 0.5 1.0

Interaction Analysis
(1) 1
(2) 2POLY
(3) .1000E 02 .2000E 00 X1
.5000E 0 1 ,5000E-01 X2
(4) 2 2
(5) 1 2
 Table 4-3
Simulation Outputf o r POLY

MODEL 1, POLY VAR. NAMES NOMINAL VALUE DEVIATION DISTRIBUTION

x1 .10000E 2 .200000E 0 NORMAL
x2 .50000E 1 .500000E -1 NORMAL
INPUT CORRELATIONS

,500

INPUT CHECK
MODEL 1, POLY VAR. NAMES NOMINAL VALUE DEVIATION DISTRIBUTION
x1 .99866E 1 .208713 0 NORMAL
x2 .50019E 1 .65881E -1 NORMAL
UI INPUT CORRELATIONS
W
,608

DEPENDENT DATA LISTED IN ASCENDING

ORDER
I IIN POLY
1 .010 .6322E 3 39 .390 .6735E
77 3. .770 .6962E 3
2 .020 40
.6346E 3 .400 .6740E 3 78 .780 .6981E 3
3 .030 .6362E 3 41 .410 .6740E 3 79 .790 .6983E 3
4 .040 .6389E 3 42 .420 80 3
.6748E .800 .6986E 3
5 .050 .6403E 3 43 .430 .6750E 3 81 .810 .7012E 3
6 .060 .6434E 3 44 .440 .67 55E
3 82 .820 .7013E 3
7 .070 .6478E 3 45 .450 .6758E 3 83 .830 .7021E 3
8 .080 46
.6494E 3 .460 .67643 3 84 .840 .7024E 3
9 .090 .6504E 3 47 .470 .6774E 3 85 .850 .70393 3
10 .1.00 .6516E 3 48 .480 .6786E 3 86 .860 .7040E 3
11 .no .6554E 3 49 .490 .6786E 3 87 .870 .7051E 3
12 .120 .6567E 3 50 .500 .6788E 3 88 .880 .7056E 3
13 .130 .6579E 3 51 .510 .6790E 3 89 .890 .7062E 3
14 .140 .6580E 3 52 .520 .6791E
90 3 .goo .7067E 3
15 .150 .6594E 3 53 .530 .67993
91 3 .910 .70963 3
F

T a b l e 4-3 (Continued)

16 .160 .6607E 3 54 ,540 .6817E 3 92 .920 .7099E

.170 17 .6612E
55 3 .550.930 .6820E
93 3 .7113E
18 .180 ,66143 3 56 .940
.560 94
.6833E 3 .7117E
,190 19 .6615E 3 57 .570 3
.6851E .950 95 .7125E
.200 20 .6615E
58 3 .580 3
.6856E .960 96 .7147E
.210 21 .6634E
59 3 .590 .6858E 3 97 .970 .7158E
22 ,220 .6637E
60 3 .600.980 98
.6858E 3 .7167E
,230 23 ,66473 3 61 .610 3
.6860E .990 99 .7272E
24 .240 .6650E 3 62 .620 .6860E 3 100 1.000 .72.88E
.250 25 63
.6659E 3 .630 .6863E 3
26 .260 .6660E 3 64 .640 .6864E 3
27 .270 .6661E 3 65 .650 .6872E 3
.280 28 66
.6676E 3 .660 .6882E 3
29 .290 .6678E 3 67 .670 .6892E 3
.300 30 68
.6692E 3 .680 .6893E 3
31 .310 69
.6693E 3 .690 .6893E 3
32 .320 .6696E 3 70 .700 .6903E 3
33 .330 .6704E 3 71 .710 .6904E 3
.340 34 .6706E 3 72 .720 .6906E 3
.350 35 .6711E 3 73 .730 .6933E 3
36 .360 .6715E 3 74 .740 .6933E 3
37 .37Q$ 75
.6726E 3 .750 .6951E 3
38 .380 .6734E 3 76 .760 .6953E 3

MOMENTS POLY PERCENTAGE

POINTS
FOR POLY BY EDGEWORTH
FIRST .680057E
3 Z = 616.96093 F(Z) = -.82690E -2

SECOND .437902E
5 Z = 627.47654 F(Z) = -.11345E -1
THIRD -.243477E
5 Z = 637.99217 F(Z) = .88139E -2
FOURTH .503237E 8 Z = 648.50779 F(Z) = .83527E -1
STD. DEV. .210316E
2 Z = 659.02342 F(Z) = .21687E 0
SKEWNESS -.265701E -1 Z = 669.53905 F(Z) = .36661E 0
KURTOSIS .262433E -1 Z = 680.05467 F(2) = .49822E 0
VARIANCE - COVARIANCE MATRIX, ORDER 1 Z = 690.57030 F(Z) = .63100E 0
i

Table 4-3 (Continued)

POLY .468040 E3
Z = 701.98592 F(2) = .78309E 0
z = 711.60155 F(Z) = .917883 0
Z = 722.11718 F(Z) = .99259E 0
z = 732.63280 F(Z) = .10121E 1
Z = 743.14842 F(Z) = .10086E 1
 Table 4-4
S e n s i t i v i t y , Worst-case, and Moment Analysis Output f o r POLY

9 I1
FIRST AND SECOND PARTIALDERIVATIVES (Y AND Y ) OF POLY WITH RESPECT TO X
PARTIALS SENSITIVITY
1 11

X Y (X-2DX) Y (X-1DX) Y(X+lDX) Y (X+2DX) Y Y LINEAR NDN-LIN

X1 .64284E 3 .66175E 3 .70055E 3 .72043E 3 .96986E
.79590E
2 1 56967E -1 . 9 3 4 9 9 E -3
X2 .67384E 3 .67740E 3 .68460E 3 .68823E 3 .71995E 2 .78125E 1 . 1 0 5 7 2 E -1 . 5 7 3 6 1 E -4
ALL X AT NOMINAL, Y(X) = .68099E 3
STD DEV OF Y ( X ) , .21425E
2

WORST CASE LIMITS

VALUE OF VARIABLE AT
LOWER LIMIT AND
AT UPPER LIMIT , X DX
x1 .96000E 1 .10400E 2 .10000E 2 .20000E 0
x2 .49000E 1 .51000E 1 .50000E 1 . 5 0 0 0 0 E -1

WORST CASE LIMITS AND

NOMINAL VaUE
POLY .63579E 3 .72779E 3 .68099E 3

INTERACTION CHECK USING 1 S T AND 2NDDEGREETERMS OF TAYLOR SERIES

POLY .63567E 3 .72766E 3
INTERACTION CHECK USING 1 S T DEGREETERMS OF TAYLOR SERIES
POLY .63500E 3 .72698E 3
GOODNESS OF F I T USING 1ST AND 2NDTERMS OF TAYLOR SERIES
VARIABLES Y(X-2DX) /Y (X) l.-SENS 1.-SENS+NON LIN Y (X+2DX) /Y (X) 1 .+SENS l.+SENS+NON LIN
x1 .94397E 0 .94303E 0 .94397E 0 .lo5793 1 .10570E 1 .lo5793 1
.98949E x2 0 .98943E 0 .98948E 0 .10106E 1 .10106E 1 .10106E 1
 Table 4-5
Interaction Analysis Output for POLY

VARIABLE
NOMINAL
DX
VALUE

x1 .10000E 2 ,20000E 0
.50000E x2 1 .50000E -1

CODED LEVELSOF THE VARIABLESX(1)

0-LOWLEVEL1-HIGHLEVEL

MOD-2
ROW
ARRAY
VARIABLES
OF

1 0 0
2 0 1
3 1 0
4 1 1

ACTUAL LEVELSOF X(1) AND CORRESPONDING PERFORMANCE VALUES

ROW x1
POLY x2

1 .96000E 1 .49000E 1 .63579E

3
.96000E
2 1 .51000E 1 .64995E
3
.10400E
3 .49000E
2 1 .71316E
3
4 .10400E .51000E
2 1 .72779E
3

COEFFICIENTS OF VARIABLES AND THEIR SENSITIVITIES

COEFFICIENTS
SENSITIVITY

CONSTANT B( 0) = .68167E
3
x1 B( 1) = .56938E
.96938E 2 -1
x2 B( 2) = .lo4963
.71480E 2 -1
x1 , x2 B( 1, 2) = .29087E 1 ,854233 -4
4.3.2 ECAJ? and NASAP f o r PVA
The E l e c t r o n i c C i r c u i t A n a l y s i s P r o g r a m (ECAP) w a s d e v e l o p e d j o i n t l y by IBM
andNorden DivisionofUnitedAircraft; Ref.4-16 is t h e b a s i c r e f e r e n c e f o r t h e
program. ECAP is v e r yw i d e l yu s e df o rc i r c u i ta n a l y s i s ; i t i s a v a i l a b l ef r o m IBM
for use on the IBM 1620,7000 series, and360 series c o m p u t e r s , a l t h o u g h n o t a l l o f
these v e r s i o n s are o f f i c i a l l y s u p p o r t e d by IBM [Ref.4-171. It h a sb e e ns u i t a b l y
modified by o t h e r o r g a n i z a t i o n s f o r u s e on a v a r i e t y o f o t h e r c o m p u t e r s a n d w i t h
some v a l u a b l e a d d i t i o n a l f e a t u r e s f o r PVA.
In the versions of ECAF' a v a i l a b l e . f r o m IBM, t h e PVA c a p a b i l i t i e s i n c l u d e t h e
f o l l o w i n g [Ref.4-181:
For dc analysis:
(1) partial derivative o f voltage a t a p a r t i c u l a r c i r c u i t nodewithrespect
t o a circuit parameter in a particular branch;
(2) sensitivity of a node voltage with respect to a branchparameter;
( 3 )w o r s t - c a s es o l u t i o n s ;
( 4 )s t a n d a r dd e v i a t i o n of c i r c u i to u t p u tv a r i a b l e s ;
(5) a u t o m a t i c p a r a m e t e r v a r i a t i o n , w h i c h a l l o w s a p a r a m e t e rt ob ei n c r e m e n t e d
over a r a n g e o f v a l u e s w i t h a circuit solution computed f o r e a c h v a l u e .
For a c a n a l y s i s :
( 1 )a u t o m a t i cp a r a m e t e rv a r i a t i o n .
A d d i t i o n a l PVA c a p a b i l i t i e s w h i c h h a v e b e e n i n c o r p o r a t e d i n ECAP by o t h e r o r g a n i z a t i o n s
i n c l u d e ac s e n s i t i v i t i e s and s o l u t i o no ft h ep r o p a g a t i o n - o f - v a r i a n c ee q u a t i o n[ R e f . 4 - 1 7 ] ,
TheNetwork Analysis or System Application Program (NASAP) has been developed
byNASA/ElectronicsResearchCenterin a cooperativeeffortinvolvingabout 20 u s e r s
of t h e program [Ref .4-191. NASAP is unique among c i r c u i t a n a l y s i s p r o g r a m s i n t h a t
i t u s e sf l o w g r a p ht e c h n i q u e st oa n a l y z en e t w o r k si n s t e a do fm a t r i x - o r i e n t e dt e c h n i q u e s .
Also, i t m a n i p u l a t e s c i r c u i t s y m b o l i c p a r a m e t e r s i n s t e a d o f a c t u a l p a r a m e t e r s u n t i l
thefinalstep o ft h ea n a l y s i s .T h i ss y m b o l - m a n i p u l a t i o nf e a t u r eh a s some i n t e r e s t i n g
r a m i f i c a t i o n s , among which are t h e a b i l i t y t o c a l c u l a t e p a r t i a l d e r i v a t i v e s and
s e n s i t i v i t i e ss y m b o l i c a l l y [Ref.4-201.
In addition to the PVA c a p a b i l i t i e s n o t e d a b o v e , NASAP i n c o r p o r a t e s an optimi-
zationprocedurewhicheliminates from a c i r c u i t i n p u t p a r a m e t e r s h a v i n g less t h a n
a p r e a s s i g n e d amount o f i n f l u e n c e on c i r c u i t p e r f o r m a n c e p a r a m e t e r s ; t h e p r o c e d u r e
is i n e f f e c t a tolerance analysis [Ref .4-201.
NASAP was o r i g i n a l l y w r i t t e n i n FORTRAN IV f o r u s e on t h e CDC 3600computer;
i t a l s o is now i n u s e on s e v e r a lo t h e rc o m p u t e r s .A l t h o u g hr e p o r t e d l ya v a i l a b l e
from COSMIC [Ref.4-21], it doesnotappearintheJuly1967listingof COSMIC

58
programs[Ref.4-221. However, i t c a nb eo b t a i n e d [Ref.4-23] by c o n t a c t i n g :

R. M. C a r p e n t e r
NASA/ERC
575 Technology Square
Cambridge , Mass.
T e l . 617 491-1500, Ext.541

References

4-1. Milne, W.E.: Numerical C a l c u l u sP. r i n c e t o U

n niversity Press, Princeton,
N. J . , 1949.

4-2. S t a f fo fA u t o n e t i c sD i v .o fN o r t h American A v i a t i o n :R e l i a b i l i t yA n a l y s i so f
E l e c t r o n i cC i r c u i t s .( A v a i l a b l e from DDC as AD 461303).

4-3. Nelson, A. C.; e t . a l . : Developmentof R e l i a b i l i t y Methodology f o r Systems

Engineering,Vols. I and 11. R e s e a r c hT r i a n g l eI n s t i t u t e ,R e s e a r c hT r i a n g l e
Park,NorthCarolina, (NASA C o n t r a c t NASw-905), A p r i l 1966.

4-4. Abramowitz, M.; and Stegun, I. A.: Handbook ofMathematicalFunctionswith

Formulas , Graphs, and MathematicalTables. Dover P u b l i c a t i o n s , New York,
New York,1965.

4-5. Dorf,Richard C.: Computer-Analyze

Your C i r c u i t . .E l e c t r o n i cD e s i g n ,V o l . 13,
June21,1967,pp. 54-57.

4-6. O'Bryant, R. 0.: V a r i a b i l i t yP r e d i c t i o n - - A New Method. Proceedings1967

Symposium on R e l i a b i l i t y ,W a s h i n g t o n , D. C . , pp.181-188.

4-7 Freund, J . E.: M a t h e m a t i c aSl t a t i s t i c sP. r e n t i c e - H a l lI,n c . , Englewood C l i f f s ,

N. J . , 1962.

4-8. K i e f e r , F. P . , e t . a l . : FinalReport on P r e d i c t i o n of C i r c u i tD r i f tM a l f u n c t i o n
of S a t e l l i t e Systems,Report ARPA 168-61. IBM, FSD SpaceGuidanceCenter,
Owego, New York f o r Rome A i r Development C e n t e r , G r i f f i s s Air Force Base,
C o n t r a c t AF 30(602)-2418, AD 276 044,1961.

4-9. Hayward, R. A.; and I n g b e r , J . I.: Monte C a r l oF l i g h tP e r f o r m a n c eR e s e r v e

Program.
Report GD/C-BTD-65-176. G e n e r a l Dynamics ConvairDivision, STAR
N66-22903, 1966.

4-10. Markowitz, H. N.; e t . a l . : Launch V e h i c l eO p t i m i z a t i o n - Phase I1 F i n a lR e p o r t ,

Vol. I1 -
TechniquesDevelopment. Lear S i e g l e r ,I n c . ,C o n t r a c t NASw-398.
1965 , STAR N66-14951.

4-11. B o s i n o f f , I.; e t . a l . : T r a n s f e rF u n c t i o n si nM a t h e m a t i c a lS i m u l a t i o nf o r
R e l i a b i l i t yP r e d i c t i o n ,F i n a lR e p o r t ,R e p o r t RADC-TDR-63-87, Sylvania
E l e c t r o n i cS y s t e m s Command, C o n t r a c t AF 30(602)-2376,1963.

4-12. K e n d a l l , M. G.: TheAdvanced Theoryof S t a t i s t i c s , Vol. 1, C h a r l e sG r i f f i n and

Co., Ltd., London,England,1948.
References(Continued)

4-13. West, W. J . ; a n dS c h e f f l e r , H. S.: D e s i g nc o n s i d e r a t i o n sf o rr e l i a b l ee l e c t r o n i c

equipment.NortheastResearch and EngineeringMeeting,Boston, Mass., 1961.

4-14. M a r i n i ; Brown; and Williams: E v a l u a t i o na n dp r e d i c t i o n of c i r c u i tp e r f o r m a n c e

by s t a t i s t i c a l t e c h n i q u e s .A e r o n a u t i c a lR a d i o ,I n c . , Monograph No. 5 ,
P u b l i c a t i o n No. 113.

4-15. Addelman, S . : T e c h n i q u e sf o rc o n s t r u c t i n gf r a c t i o n a lr e p l i c a t ep l a n s , J.
Am.
S t a t i s .A s s o c . ,5 8 , 45-71.

4-16. IBM T e c h .P u b l i c a t i o n sD e p t . :1 6 2 0E l e c t r o n i cC i r c u i tA n a l y s i sP r o g r a m (ECAP)

(1620-EE-02X). IBM P u b l i c a t i o n H20-0170-1, W h i t eP l a i n s , N. Y . , 1965.

4-17. Wall, H. M.: ECAP-1966. NEREM Record-1966,

pp.84-85.

4-18.Tyson , H. N . , J r . ; H o g s e t t , G. R . ; andNisewanger, D. A.: The I B M E l e c t r o n i c

C i r c u i tA n a l y s i sP r o g r a m (ECAP). Proceedings1966Annual Symposium on
R e l i a b i l i t y , pp. 45-65.

4-19. C a r p e n t e r , R. M.: NASAP--Network AnalysisforSystemApplicationProgram--Present

C a p a b i l i t i e s o f a MaintainedProgram. Computer-Aided C i r c u i tD e s i g nS e m i n a r
P r o c e e d i n g s ,A p r i l 11-12, 1967,KresgeAuditorium, MIT, pp. 83-94.

4-20. C a r p e n t e r , R. M. ; and Happ, W . : Computer-Aided Design--Part 3 A n a l y z i n gC i r c u i t s

w i t hS y m b o l s E
. lectronics, Dec. 12,1966,pp. 92-98.

4-21. C a r p e n t e r , R. M.: TopologicalAnalysisofActiveNetworks.Proceedingsof

theInstituteon Modern S o l i d S t a t e CircuitDesign,Univ. of S a n t a Clara,
C a l i f o r n i a ,S e p t1 9 6 6 ,p p . 50-59.

4-22. Computer Center,Univ.ofGeorgia: A D i r e c t o r yo f ComputerPrograms Available

from COSMIC, Vol. 1, J u l y 1, 1967.

4-23. Dumanian, J. A., ed.: I E E E CADAR Newsletter. January1967.

60
5. P a r t A p p l i c a t i o nA n a l y s i s
In part application analyses the operating stresses o f t h e i n d i v i d u a l c o m p o n e n t s
are determinedandcompared t ot h er a t e dc a p a b i l i t i e s .I n an e l e c t r o n i c c i r c u i t , f o r
example, p a r t stresses such as power d i s s i p a t i o n o f a resistor, peak reverse voltage
of a d i o d e , and v o l t a g e a c r o s s a c a p a c i t o r are a l l t a b u l a t e d a n d compared t o t h e i r
electrical r a t i n g s . Theconceptof stress h e r e is a ne x t e n s i o n of t h ec o n c e p t of
mechanical stress a p p l i e d i n s t r e n g t h o f materials a n a l y s i s a n d i s broadened t o i n c l u d e
electrical, thermal, radiation and other potentially damaging e f f e c t s t h a t may j e o p a r d i z e
t h ea c c e p t a b l eo p e r a t i o n of a component. The p u r p o s eo ft h ea n a l y s i s is t o i n s u r e t h a t
a c t u a l component l o a d s do n o t e x c e e d t h e m a n u f a c t u r e r ' s r a t e d o r u s e r ' s d e r a t e d c a p a -
b i l i t y o ft h e component.
The s i g n i f i c a n t a p p l i c a t i o n o f c o m p u t e r s i n p a r t a p p l i c a t i o n a n a l y s i s is i n d i r e c t l y
t h r o u g ho t h e rt y p e s of a n a l y s e ss u c h as c i r c u i t , t h e r m a l and s t r u c t u r a la n a l y s e s .F o r
e x a m p l e ,w i t hc i r c u i ta n a l y s i sp r o g r a m ss u c h as ECAP, nodevoltages and b r a n c h c u r r e n t s
(hencebranch componentpower dissipation)ofelectricalnetworkscanbe computed f o r
l a t e r comparison t or a t e dc o n d i t i o n s . The c i r c u i ta n a l y s i s program NET-1 a l l o w sa s
i n p u tt h er a t e dd cc o n d i t i o n s of c e r t a i n components,performs a comparisonagainst
ratedvalues as a p a r t of t h e a n a l y s i s , and p r i n t s o u t a n alarm i f a computed parameter
v a l u ee x c e e d st h ei n p u tr a t e dv a l u e .M e c h a n i c a l stress a n a l y s i s i s u s u a l l ya ni n h e r e n t
featureinstructural:analysisprograns,sincethe stress l e v e l i n a s t r u c t u r e i s con-
c e r n e dw i t ht h ep r i m a r yf u n c t i o n of t h e s t r u c t u r e .
The c o m p u t e r c a n s e r v e a s a n a i d t o a p p l i c a t i o n a n a l y s e s on systemcomponents
' f o r any s i t u a t i o n i n which t h e component loadscanbe computed w i t h a n a p p r o p r i a t e
model.Vol. V of t h i s r e p o r t series t r e a t s p a r t a p p l i c a t i o n a n a l y s e s i n some d e t a i l .

61
6. F a i l u r e Mode and E f f e c t sA n a l y s i s (J?MEA)
T h i sa n a l y s i st a s k is approached i n s e v e r a l ways.The common purposeof all
approaches i s t o d e t e r m i n e w h a t d i s c r e p a n c i e s c a n o c c u r i n a system,identifytheir
e f f e c t s onsystemoperation,andeliminatethosethat are more c r i t i c a l andmore
l i k e l yt oo c c u r . A l a r g ep o r t i o no ft h ea n a l y s i s relies onengineeringjudgement
and is thusperformedmanually.Computerscan assist, b u tt h ee x t e n t of a p p l i c a b i l i t y
depends on t h ea p p r o a c ht a k e na n dt h en a t u r eo ft h es y s t e m . FMEA r e m a i n st h ei m p o r t a n t
p r o c e d u r ef o ra c t u a l l yu n c o v e r i n gt h es y s t e md i s c r e p a n c i e s . I t is i n f a c t one of t h e
mostimportantactivitiesinthetotal.designforreliabilityprocesssince it identifies
areas r e q u i r i n ga c t i o nb yo t h e rd e s i g na c t i v i t i e s . One of i t s i m p o r t a n to u t p u t s is
thedesignationofthelogicmodelsforindividualelementstobeincludedinreliability
prediction calculations.
One of t h es i m p l e s ta p p r o a c h e st o FMEA i s : g i v e n a d e s i g nc o n f i g u r a t i o n ,e a c h
of t h e componentsand materials c o m p r i s i n g t h e d e s i g n c a n f a i l or degrade via a number
of d i f f e r e n t modes.The f a i l u r e mode a n a l y s i s c o n s i s t s o f n o t h i n g more t h a n e x p l i c i t l y
i d e n t i f y i n gt h e s e modes. For a system composed of d i s c r e t e components, t h i s i d e n t i f i -
c a t i o ni n v o l v e sm e r e l yp r o c e e d i n gt h r o u g h a p a r t s l i s t and decidingwhat modes of
f a i l u r e are t ob ec o n s i d e r e d .T h e r e is a p r a c t i c a l l i m i t , of c o u r s e , as t o how
many
f a i l u r e modes o f e a c h p a r t c a n b e c o n s i d e r e d , a n d i n fact a limited failure effects
a n a l y s i s is performed on a s u b j e c t i v e b a s i s at t h i s s t a g e t o a i d i n l i m i t i n g t h e
number of modes c o n s i d e r e d .
Forelectroniccircuits i t i s becoming f a i r l y common t o c o n s i d e r a t least s h o r t s
andopensbetween a l l t e r m i n a lp a i r s of components.Typical modes d e f i n et h ee x t r e m e
d i s c r e t e s t a t e s o ft h e components. It i s p o s s i b l et od e f i n ei n - b e t w e e n s t a t e s , such as
discrete levels of resistance for a r e s i s t o r which d i f f e r fromnominal,butthe re-
sultinganalysiscanquickly become unweildy i f c a r r i e d t o o f a r , e s p e c i a l l y when
c o n s i d e r i n gd e v i c e s as complex as a t r a n s i s t o r .
When f o r FMEA t h e l o w e s t l e v e l o f breakdown is l i m i t e d t o complex s u b a s s e m b l i e s
(such as t r a n s m i t t e r s , power i n v e r t e r s , pumps, and e n g i n e s ) t h e f a i l u r e modes become
much more d i f f i c u l t t o d e f i n e .I ft h e s es u b a s s e m b l i e s are r e q u i r e dt op e r f o r mi n
sequencesofoperations,failure modes o f t h e f o l l o w i n g t y p e s may b e i d e n t i f i e d :
( 1p)r e m a t u roep e r a t i o n ,
(2) f a i l u r et oo p e r a t e a t a p r e s c r i b e d time,
(3) f a i l u r et o cease o p e r a t i o n a t a p r e s c r i b e d time, and
(4) f a i l u r ed u r i n go p e r a t i o n .
Withineach of t h e s e modes t h e r e may b e f u r t h e r modes t oc o n s i d e r .F o re x a m p l e ,
f a i l u r e of a power s u p p l y d u r i n g o p e r a t i o n may be evidenced by e i t h e r no output

62
voltage,lossofvoltageregulation,frequencyout of t o l e r a n c e , o r e x c e s s i v e v o l t a g e
imbalance between different phases.
The o n l y a i d p r o v i d e d by a computer i n the f a i l u r e mode p o r t i o n of t h e a n a l y s i s
is t h a to fr e c o r dk e e p i n gt oe l i m i n a t em a n u a ld r u d g e r y .T h i sr o l e becomesmore useful
when t h e r e c o r d s c a n b e u s e d as i n p u t t o t h e f a i l u r e e f f e c t s a n a l y s i s , which p o t e n t i a l l y
lends itself to morecomputer assistance.
The u s e o f t h e c o m p u t e r i n t h e f a i l u r e e f f e c t s p o r t i o n of t h e a n a l y s i s is
primarily in the role of function evaluation using performance models t o compute
changes i n p e r f o r m a n c ed u et op a r t i c u l a rf a i l u r e modes. Forexample, considerations
of f a t i g u e f a i l u r e of a p a r t i c u l a r s t r u c t u r a l member w i l l n o t a l t e r t h e b a s i c f o r m
ofthestiffnessmatrixbut w i l l m o d i f yt h ev a l u eo fc e r t a i np a r a m e t e r s . Upon sub-
stituting the modified values into the computerprogram for solving these structural
e q u a t i o n s ,t h ec o m p u t e rc a nb eu s e dt oe v a l u a t et h ee f f e c t .
It is p o s s i b l e t o e x t e n d c e r t a i n p e r f o r m a n c e e v a l u a t i o n p r o g r a m s t o a u t o m a t i c a l l y
performthesecalculationsforallfailure modes t o b e i n v e s t i g a t e d . The NET-I network
a n a l y s i s program[Ref. 6-11 d o e s t h i s upon i n p u t r e q u e s t f o r a l i m i t e d number of ab-
normal modes o f c i r c u i t v o l t a g e s u p p l i e s and p r i n t s o u t t h e v a l u e o f c i r c u i t p e r f o r -
mance p a r a m e t e r sf o re a c h . NET-I d o e sn o ta u t o m a t i c a l l yc o n s i d e rf a i l u r e modes such
as s h o r t s andopens of c i r c u i t components; i n v e s t i g a t i o n of t h e s e would r e q u i r e
manually setting upa new r u n t o b e made f o r e a c h mode.
Most c i r c u i at n a l y s i s programs
e.g., ECAP which a c c e p t a t o p o l o g i c ailn p u t
description of the circuit and s y n t h e s i z e t h e c i r c u i t e q u a t i o n s c a n b e u s e d t o e v a l u a t e
failureeffects,but computerrun t i m e can become e x c e s s i v e s i n c e t h e c i r c u i t e q u a t i o n s
may have t o b e g e n e r a t e d a g a i n f o r e a c h r u n . S p e c i f y i n g a n e x t r e m e f a i l u r e mode such
as anopen o r a s h o r to f acomponent e s s e n t i a l l yc h a n g e st h ec i r c u i tc o n f i g u r a t i o n
and a completely new s o l u t i o n is r e q u i r e d . A u s e f u la p p r o x i m a t i o nt o open o r s h o r t
failures often used is to maintain the same c i r c u i t c o n f i g u r a t i o n and merely use
e x t r e m e l yh i g ho r low v a l u e s o f p a r t p a r a m e t e r s t o s i m u l a t e f a i l u r e s . F o r example,
an e x t r e m e l y h i g h c a p a c i t a n c e v a l u e c a n e f f e c t i v e l y s i m u l a t e a s h o r t of a c a p a c i t o r
f o r AC a n a l y s i s b u t d o e s n o t h a v e t h e same e f f e c t on c i r c u i t e q u a t i o n s as does a
short.
The AMAP (Automated F a i l u r e Mode A n a l y s i s Program) c i r c u i t a n a l y s i s program
[Ref. 6-21 is o n e p r o g r a m w h i c h a u t o m a t e s t h e f a i l u r e e f f e c t a n a l y s i s f o r d c c i r c u i t s .
It r e p e a t e d l y s o l v e s t h e c i r c u i t e q u a t i o n s , computingand printing circuit node
voltages,forfailure modes such as openand short for parts and s h o r t s between a l l
n o d ep a i r s . As d e s c r i b e d i n t h e r e f e r e n c e , AMAP i n c l u d e so n l yr e s i s t o r s ,d i o d e s ,
t r a n s i s t o r s , power s u p p l i e s andnodes.Thisautomatedapproach t of a i l u r ee f f e c t s

63
analysiscancarryovereffectivelyinothertypesofsystemssuch as s t r u c t u r e s and
propulsion, but no programs are known w h i c h p r o v i d e t h e s e c a p a b i l i t i e s .
A s mentioned e a r l i e r , t h e r e i s a p r a c t i c a l l i m i t t o t h e number o f f a i l u r e modes
ofeach component o r material thancanbeconsidered,evenwithcomputers. As a
r e s u l t , most f a i l u r e e f f e c t s a n a l y s e s are l i m i t e d t o f i r s t - o r d e r e f f e c t s , i.e., to
consideringtheeffect of a s i n g l e f a i l u r e mode ofonecomponent a t a time andignoring
combinations. The A" p r o g r a md o e si n c l u d es e c o n do r d e re f f e c t st o a l i m i t e de x t e n t ,
i n c l u d i n go p e na n ds h o r tc o m b i n a t i o n sb e t w e e nd i f f e r e n tt e r m i n a lp a i r so f a transistor.
One of t h e m a j o r u s e s o f t h e o u t p u t s of FMEA i s i d e n t i f i c a t i o n of t h e models
f o ri n d i v i d u a le l e m e n t st ob eu s e di n a r e l i a b i l i t yp r e d i c t i o na n a l y s i s .F o re x a m p l e , ,

theresultscanbeusedtodecidewhether a short or a particularresistorshouldbe

i n c l u d e di nt h ep r e d i c t i o n as a f a i l u r e o r s u c c e s s f u l o p e r a t i o n . A n o t h e r u s e is t o
a i di nd e t e r m i n i n gi ft h e r e are any o v e r s t r e s s c o n d i t i o n s on c i r c u i t p a r t s . The
i d e n t i f i c a t i o n of f a i l u r e e f f e c t s a l s o a s s i s t s i n c o m p i l i n g a failuredictionaryto
beused infaultdiagnosis and t e s t p o i n t a l l o c a t i o n .
Anotherapproach t o FMEA i s t o a p p l y t h e a b o v e p r o c e d u r e i n r e v e r s e , i.e., to
d e f i n e a degradedorfailed mode o f t h e s y s t e m and l o o k f o r t h o s e componentand
material f a i l u r e s t h a t c a n c a u s e it. The approach i s employed m a i n l yf o rs t u d y i n g
t h em i s s i o ns e q u e n c e so ff u n c t i o n sf o rl a r g es y s t e m s . I n t h i sa p p r o a c ht h e ex-
c i t a t i o no r" c a l l i n g - u p "o f a functiondepends on t h e mode of o p e r a t i o n of a
f u n c t i o ni n a p r i o r time i n t e r v a l . T y p i c a l l y t h e r e i s o n l yo n ep a t ht h r o u g ht h e
n e t w o r kf o rn o r m a lo p e r a t i o n . Any o t h e r p a t h c o r r e s p o n d s t o d e g r a d a t i o n o r f a i l u r e

F i g u r e 6-1. FixedSystemwith Redundancy

64
and w i l l e x c i t e f u n c t i o n s w h i c h c a u s e t h e m i s s i o n t o b e a b o r t e d , f a i l e d , o r c o m p l e t e d
i n a degraded mode. Thus g i v e n a p a r t i c u l a r outcome o r t e r m i n a t i n g mode o r t h e
s y s t e m ,t h ea n a l y s i sc a ns e a r c ho u tt h o s ee v e n tc o m b i n a t i o n st h a tc a nl e a dt o it.
A thirdapproachto FMEA is u s e f u l f o r s y s t e m s w i t h f i x e d c o n f i g u r a t i o n s and
c o n t a i n i n ge x t e n s i v er e d u n d a n c y .C o n s i d e rt h ec o n v e n t i o n a ls y s t e ml o g i cd i a g r a m shown
i n F i g . 6-1. A first-order FMEA performedfrom t h i s diagram is t r i v i a l s i n c e i t was
r e q u i r e dp r i o rt od i a g r a mc o n s t r u c t i o n anyway. A second-order FMEA shows t h a t com-
b i n a t i o n ss u c h as elements A and B andelements D , F, and H c a u s e s y s t e m f a i l u r e .
When theredundancygetsverycomplex,thecomputercan assist i n p e r f o r m i n g t h e
h i g h e r - o r d e r FMEA.
References
6-1. Malmberg, A . F.: NET-1 Network AnalysisProgram.Proceedings1965 Symposium
on R e l i a b i l i t y and Q u a l i t yC o n t r o l ,p p . 510-517.

6-2. H a u s r a t h , D. A . ; and R a n a l l i , R . : Computer S t u d i e s o f AbnormallyOperating

C i r c u i t s .P r o c e e d i n g s1 9 6 6 Annual Symposium on R e l i a b i l i t y , pp. 66-86.

65
7. R e l i a b i l i t yP r e d i c t i o n
A basic definition of the reliability of an equipment i s t h e p r o b a b i l i t y t h a t t h e
equipmentsuccessfullyperforms its intended function for a specified duration.while
o p e r a t i n gu n d e rc e r t a i ne n v i r o n m e n t a lc o n d i t i o n s .R e l i a b i l i t yp r e d i c t i o n i s t h ep r a c t i c e
of u s i n g m a t h e m a t i c a l m o d e l s t o estimate t h i s p r o b a b i l i t y o r r e l a t e d m e a s u r e s s u c h as
p r o b a b i l i t yo ff a i l u r e ,l i f ed i s t r i b u t i o n s ,o rm e a n - t i m e - t o - f a i l u r e .I na d d i t i o nt o
t h e s e estimates o f s y s t e m r e l i a b i l i t y a l o n e , p r e d i c t i o n o f more complexmeasuresof
s y s t e mw o r t hr e l a t e dt or e l i a b i l i t yc a nb e made. Forexample, i t may b e d e s i r e d t o
optimize system reliability under cost'constraints; a computerprogramwhichaccomplishes
t h i so p t i m i z a t i o n i s d i s c u s s e dl a t e r .R a r e l ya r et h em o d e l so r statistics s u f f i c i e n t
toobtainanestimatewithsufficientaccuracytohavemeaningintheabsolutesense.
However, t h e r e s u l t s do f r e q u e n t l y h a v e meaning a s a b a s i s f o r s e l e c t i n g t h e b e s t of
severalcandidatedesigns, and t h e p r a c t i c e of p r e d i c t i n g s y s t e m r e l i a b i l i t y i s now
found i n almost a l l systemdevelopmentprograms.
Modelsand techniquesforpredictionaredescribedin Vol. I V - P r e d i c t i o n of
thisseries; we h e r ee m p h a s i z et h ea u t o m a t i o no ft h ep r e d i c t i o na n a l y s e s .
Reliability predictions are performedboth on i n d i v i d u a l items andon t h e combina-
t i o n s of items f o r m i n gh i g h e rl e v e l so fa s s e m b l y up t o a n d i n c l u d i n g t h e l a r g e s t of
s y s t e m s .F o ri n d i v i d u a l items t h e a n a l y s i s i s u s u a l l y so s i m p l e as t oh a v e noneed
f o r a computer.Computersdo f i n dc o n s i d e r a b l ea p p l i c a t i o ni nt h ea n a l y s i sf o r com-
bined items.
The common b a s i s f o r all reliabilitypredictions is t h e l o g i c which d e f i n e s t h e
e v e n t so fi n t e r e s t .T h i sl o g i cc o m p r i s e st h es y s t e mm o d e l ;n o ts u r p r i s i n g l y it is
c a l l e dt h ep r e d i c t i o n model. The e v e n to ft h es y s t e mb e i n gi n a particularstate
(insimplestformthestate is eithersuccessorfailure) is t h e l o g i c c o m b i n a t i o n
ofothereventsassociatedwith states o f s y s t e m s u b a s s e m b l i e s , i n p u t s t o t h e s y s t e m ,
l o a d s on t h es y s t e m ,a n d / o rs y s t e me n v i r o n m e n t a lc o n d i t i o n s .I nc o n c e p tt h el o g i c
comprising the preciction modelcanallowany number o f d i f f e r e n t s t a t e s of a p a r t ;
most a n a l y s e s of. complexequipmentemploy s i m p l e two-statemodels(successvs.
f a i l u r e )t o limit a n a l y s i sc o m p l e x i t y .
The b a s i c f l o w o f p r o c e d u r e s i n r e l i a b i l i t y p r e d i c t i o n i s shown i n Fi g. 7-1.
A major milestone i s t h e p r e d i c t i o n model,fromwhich either of two b a s i c a p p r o a c h e s
may b ef o l l o w e d .T h ea p p r o a c hi l l u s t r a t e d by t h e u p p e r p a t h l e a d s t o a prediction
equationwhichexpressestheprobabilityofsystemsuccessor a relatedmeasure as
a f u n c t i o no fi n d i v i d u a le l e m e n tp r o b a b i l i t i e s . One of t h es i m p l e s tr o l e s of com-
puters in reliability prediction i s tousesuchanequation programmed f o r e s t i m a t i n g
s y s t e m p r o b a b i l i t i e s andcomputing s e n s i t i v i t i e s ofsystemprobabilitiestochanges

66
 I
Generate
Prediction
Equation
- Compute
Solutions(s) t o
Prediction
Equation
-

>
1
Preliminary Develop
Evaluate
Analyses + Prediction
Results
(Includes FMEA) Model
r

Simulate
with
Prediction
Model

F i g u r e 7-1. R e l i a b i l i t yP r e d i c t i o n .P r o c e s s

i np r o b a b i l i t i e so fs u b s y s t e me v e n t s .T h i s i s e s p e c i a l l ya p p r o p r i a t e when t h ep r e -
d i c t i o ne q u a t i o n is derivedmanually and is toocomplexformanualsolution. A
computerapplicationwhichimplementsthelowerpathinthefigure is t h e u s e o f t h e
p r e d i c t i o n model f o r s i m u l a t i n g t h e s y s t e m by Monte C a r l o m e t h o d s , t o estimate t h e
p r o b a b i l i t y of s y s t e ms u c c e s so ro t h e rr e l i a b i l i t yp a r a m e t e r s .C o m p u t e r sa l s oc a n
beusedtocovervariouscombinations of t h e s t e p s i l l u s t r a t e d i n t h e f i g u r e .
7.1 Developing t h eP r e d i c t i o n Model
A prerequisitetothepredictionanalysis is a p r e l i m i n a r y a n a l y s i s of t h e e q u i p -
mentand its operational profile to establish mission functions, operating times and
sequences,andenvironments. A f a i l u r e modes and e f f e c t s a n a l y s i s as d e s c r i b e d e a r l i e r
i s an i m p o r t a n t p a r t of t h e p r e l i m i n a r y a n a l y s i s , p a r t i c u l a r l y f o r complex s y s t e m s .
An o u t p u t of t h e FMEA i s i d e n t i f i c a t i o n o f t h e l o g i c m o d e l s t o b e u s e d t o s i n g l e e l e m e n t s
i n t h e model f o r t h e p r e d i c t i o n a n a l y s i s , which is n e x te s t a b l i s h e d . The g o a l i s t o
o b t a i n a logicrepresentationwhich relates r e l i a b i l i t y e v e n t s of i n t e r e s t ( s u c h as
s y s t e ms u c c e s s )t ot h ee v e n t st h a tc a u s e them. T h i sl o g i cc a nb ed e v e l o p e di n two
p r i n c i p a l ways as described below.
When a system i s a f i x e d c o n f i g u r a t i o n o r when t h e e v e n t o f s y s t e m s u c c e s s d u r i n g
a particular phase of system operation is c o n c e r n e d o n l y w i t h t h e f i x e d s y s t e m

67

i
c o n f i g u r a t i o nw h i c h exists duringthisphase, a l o g i c d i a g r a m is constructedwhich
t y p i c a l l y reveals t h e v a r i o u s l o g i c e l e m e n t s o p e r a t i n g i n series o r p a r a l l e l . T h i s
diagramusually i s derivedmanuallyfromfunctionaldiagrams,schematics,special
analyses,andgeneralknowledge of s y s t e mo p e r a t i o n . Forms o f l o g i c r e l a t i o n s h i p s
thatcanbeusedinpredictionmodels are s e r i e s - p a r a l l e l d i a g r a m s , tree diagrams,
t r u t ht a b l e s , andstate-spacediagrams.Althoughcomputers are n o ts u i t e dt op r o d u c i n g
t h e p r e d i c t i o n model i t s e l f , t h e y c a n a s s i s t i n p e r f o r m i n g c e r t a i n of t h e a n a l y s e s re-
q u i r e dt od e t e r m i n ew h a tt h er e l i a b i l i t yl o g i cd i a g r a ms h o u l db e ;f o re x a m p l e ,t h e
ECAP p r o g r a m d e s c r i b e d e a r l i e r c a n b e u s e d i n FMEA.
The s e c o n d d i r e c t i o n i n w h i c h t h e m o d e l - b u i l d i n g c a n p r o c e e d is to establish the
logicrequiredtoanalyzethetotalsystemthroughoutthetotaloperationalprofile
where t h es y s t e mc o n f i g u r a t i o no rt h ee n v i r o n m e n t( o rb o t h )c a nb ec h a n g i n g . The
l o g i c mustthen relate the reliability events that occur in sequence, where each
e v e n t may r e p r e s e n t some c h a r a c t e r i s t i c o f o v e r a l l f u n c t i o n a l o p e r a t i o n o f a different
s y s t e mc o n f i g u r a t i o n .T h i st y p eo fr e p r e s e n t a t i o nl e a d s t o a cornbined f u n c t i o n a la n d
l o g i c mode; t h i s t y p e ofmodel w i l l b ec a l l e da ne v e n ts e q u e n c ep r e d i c t i o nn o d e l . The
d e v d o p u i e n to ft h i s ccuikineddiagramgenerally i s donemanually.
An e x t e n s i o n o f t h e f i r s t a p p r o a c h t o d e v e l o p i n g a p r e d i c t i o n model i s t o consid.er
t h es y s t e mr e p a i r a b l e s o t h a td i f f e r e n t s t a t e s may b ei n t r o d u c e d .T h i sl e a d st ot h e
s t a t e - s p a c ed i a g r a ma p p r o a c h ,b u th e r ea g a i n ,t h em o d e l - b u i l d i n gt a s k is primarily
a manualone.
A l l oftheaboveapproachestopredictionmodeling are d e s c r i b e d i n d e t a i l i n
vol. I V - P r e d i c t i o no ft h i s series.
7.2 Making t h eR e l i a b i l i t yP r e d i c t i o n
I t i s n o t e dh e r et h a tt h ep r e d i c t i o nc o m p u t a t i o n s a r e u s u a l l y of t h e s i m p l e r
t y p e s , i . e . ,f a i l u r e so fi n d i v i d u a le l e m e n t sa r e assumed i n d e p e n d e n t ,a n dt h ef a i l u r e
p r o b a b i l i t i e s of t h e i n d i v i d u a l e l e m e n t s a r e combined a c c o r d i n g t o t h e simple series
and p a r a l l e l l o g i c f o r a f i x e dc o n f i g u r a t i o n . An i n d i v i d u a le l e m e n tp r o b a b i l i t y is
typicallyexpressed as a d i s c r e t e p r o b a b i l i t y o r as a f a i l u r e rate w i t h a n a d j u s t m e n t
factor (called a K factor)based on theenvironment.
H a v i n gd e r i v e dt h er e l i a b i l i t yp r e d i c t i o n model, i t s u s e dependsontheapproach
t a k e nf o rp r e d i c t i o n . One much-used approach i s ( 1 )d e r i v e a Boolean a l g e b r a i c ex-
p r e s s i o nr e l a t i n gt h ee v e n t s , ( 2 ) a p p l yt h ef u n d a m e n t a l l a w s of p r o b a b i l i t y t o t h i s
expressiontoget a predictionequationwhich'expressestheprobabilityoftheout-
come e v e n t s i n terms of t h e p r o b a b i l i t i e s of t h e i n d i v i d u a l e v e n t s , and ( 3 ) apply
the prediction equation via the computer.

68
 Another approach used frequently is t o u s e t h e p r e d i c t i o n model as a b a s i s f o r
Monte C a r l os i m u l a t f o no ft h es y s t e m .T h i sr e q u i r e sa s s i g n i n ga p p r o p r i a t e numbers
t or e p r e s e n tt h ep r o b a b i l i t i e so fe a c he v e n t .F o re x a m p l ei nF i g . 7-2, t h ee v e n t A
(which i s t h e e v e n t t h a t e l e m e n t A works) i s a s s i g n e d p r o b a b i l i t y P(A) = 0.68 and
event x (the complement
of A otrh e v e ntth aet l e m e n t A d o ens ow
t o r k )p, r o b a b i l i t y b

P (x)=1-P(A) = 0.32. The comp.uter t h e n starts a p a t h s e a r c h i n g p r o c e s s s t a r t i n g w i t h

element A. A randomnumber between 0 and 1 is o b t a i n e d f r o m a random number g e n e r a t o r
routine. Thecomputer i s programmed t os e q u e n c et oe l e m e n t B if the number is less
t h a n 0.68 an8 t oe l e m e n t D i fg r e a t e rt h a n 0.68. Whicheverelement i s c a l l e d up is
treated in like manner u s i n g t h e a p p r o p r i a t e l y a s s i g n e d p r o b a b i l i t i e s f o r t h e e v e n t s
a s s o c i a t e dw i t ht h ee l e m e n t s , and thencethroughthenetwork. When a t e r m i n a le v e n t ,
e i t h e r H o r G i nt h i sc a s e , i s reached i t i s m e r e l yt a l l i e d as a h i t . Repeated t r i a l s
of t h i s p r o c e d u r e , s t a r t i n g e a c h time fromelement A, w i l l y i e l d s c o r e s f o r a l l possible
outcomes.Withenough r u n s ,t h er a t i o of t h e t a l l y f o r a p a r t i c u l a r outcome t o t h e
t o t a l number of t r a i l s w i l l p r o v i d e a n e s t i m a t e of t h e p r o b a b i l i t y of t h e p a r t i c u l a r
o u t c o m eo c c u r r i n g .T h ev a l i d i t yo ft h i se s t i m a t ed e p e n d so nt h ev a l i d i t y of t h e
numbers r e p r e s e n t i n gt h ep r o b a b i l i t yo fo c c u r r e n c eo fe a c he v e n t .T h i sa p p r o a c h is
w e l l s u i t e dt o complex systemswheresystemeventsoccur i n sequence and may represent
different system configurations.

F i g u r e 7-2. A S i m p l eP r e d i c t i o n Model

69
 A s n o r e d earlier, i t i s p o s s i b l e t o i n c l u d e morethan two states f o r e a c h e l e m e n t
used i n t h e model.Forexample, i na d d i t i o nt oc o n s i d e r i n go n l yn o m i n a la n d excess
pressureinboosterengines as e v e n t s l e a d i n g t o n o r m a l l a u n c h o p e r a t i o n and e s c a p e
towerrocketignitionrespectively, low e n g i n e p r e s s u r e m i g h t b e a n o t h e r state,
causingengineshutdown. The o n l y l i m i t o nc o m p l e x i t yo ft h ep r e d i c t i o n model is
computer s i z e andacceptablecomputing time. However, l o g i cd i a g r a m st h a ta p p e a r
simple can be deceptive in the amount ofcomputing time t h e y r e q u i r e i n p e r f o r m i n g
t h ep a t hs e a r c h i n g . S i m u l a t i o no f complex systems i s a l w a y sc o s t l y , and when many
outcomes are p o s s i b l e i t may t a k e h u n d t e d s o f r u n s t o r e a l i z e e a c h a t l e a s t once.
Checkingout a programof this type is d i f f i c u l t b e c a u s e d i s c r e p a n c i e s c a n b e d u e
e i t h e rt os y s t e ml o g i co rt ot h e program.The g u i d i n gr u l eh e r e i s t o start simple,
withonlyseveralelementstorepresentthetotalsystem, andexpand toinclude more
l o g i c d e t a i l as r e q u i r e d .
7.3 R e l i a b i l i t yP r e d i c t i o n Programs
Numerous computerprograms forreliabilitypredictionhavebeendescribedinthe
l i t e r a t u r e[ R e f s . 7-1 t o 7 - 1 4 ] ;r e l a t i v e l y l i t t l e is known a b o u t t h e i r a v a i l a b i l i t y
and s u i t a b i l i t y .T a b l e 7-1 l i s t s some of these programs. Two r e l i a b i l i t yp r e d i c t i o n
programshavebeendeveloped i nc o n n e c t i o nw i t ht h i sr e p o r tp r e p a r a t i o n .T h e s ep r o -
gramswithexamplesoftheiruses are d i s c u s s e d .
7.3.1 A Computer Program f o r System R e l i a b i l i t y
One of t h e d i f f i c u l t i e s a s s o c i a t e d w i t h o b t a i n i n g r e l i a b i l i t y estimates f o r
complex systems i s t h a t of e v a l u a t i n g p r e c i s e l y a p r e d i c t i o ne q u a t i o nw h i c he x p r e s s e s
a l l p o s s i b l ee v e n t so fi n t e r e s t . One
way toalleviatethisdifficulty is t o o b t a i n
predictionequationswhichprovideboundsonthesystemreliabilityratherthanthe
reliability itself. A method f o r d o i n g t h i s , on which t h e computerprogramgiven in
Appendix B anddiscussedbelow is based, i s developed i n Vol. I V - P r e d i c t i o n of t h i s
r e p o r t series. F o rt h ec o n v e n i e n c eo ft h er e a d e rt h a td e v e l o p m e n t i s reproducedhere.
In the l a s t fewyearsseveralpapershavebeenwritten on t h e s u b j e c t of relia-
b i l i t ya p p r o x i m a t i o n sa n db o u n d s by u s i n g t h e c o n c e p t s o f s u c c e s s p a t h s ( o r tie sets)
andcut sets. Furtherdiscussionofboundsandapproximations are given by Messinger
[Ref.7-15]. A fewof t h e more i m p o r t a n t r e s u l t s are g i v e n h e r e .
The s u c c e s sp r o b a b i l i t yo f a s y s t e m ,t y p i c a l l yc a l l e dt h es y s t e mr e l i a b i l i t y , is
d e f i n e d as t h e p r o b a b i l i t y o f s u c c e s s f u l f u n c t i o n of a l l o f t h e e l e m e n t s i n a t least
"
one -
tie set ortheprobabilitythat a l l c u t s e t s are good. A t i e s e t o rs u c c e s sp a t h
is a d i r e c t e d p a t h f r o m i n p u t t o o u t p u t as i n d i c a t e d i n t h e s i m p l e s y s t e m i n F i g . 7-3.
The t i e sets o rs u c c e s sp a t h s are 2 , 5; 1, 3 , 5; and 1, 4 , 5, respectively. A cut
-
set is a s e t of e l e m e n t s w h i c h l i t e r a l l y c u t s a l l s u c c e s s p a t h s o r t i e sets. One
i s normallyinterestedintheminimalcut s e t ; i. e., t h e smallest o r minimal s e t

70
 Table 7-1
Programs in the Reliability Prediction Area
Organizations(0riginator
Program Code Program Description or User/Sponsor) References
CRAM -Computerized Reliability Assessment Method ARINCINASA 7-1
RESCRIPT Not a specific program abut
reliability-oriented Computer Concepts,kc./? 7-2
programming language for prediction
RP-RI -Reliability Prediction of systems by combiningRadiation /?
Inc. 7-3
failure rates
RP-LG -Reliability Prediction
of systems by combining -
Lockheed-Georgia/? 7-4
failure rates
Rp-MEL -Reliability Prediction of systems by -
Marine
programmed Engineering Lab. 7-5
prediction equation
RP-G -Reliability Prediction and Crew Safety Analysis-
GrummanINASA 7-6
for complex aerospace systems from input logic
models
RP-MB -
Reliability Prediction program for computing Martin-Baltimore/?
mission - 7-7
success and crew safety for Gemini Launch Vehicle;
prediction equations required
RP-AF -Reliability.Prediction.ly simulation -
Air Force Institute of 7-8
Technology
SOAR-I1 Special purpose program for prediction of Apollo
GE-Tempo/NASA 7-9
mission success by simulation
RAPID -Reliability Analysis and Erediction Lear Siegler/NASA
Independent
of Distributions
ARM4 -Automatic Reliability Mathematical Model NAA/? 7-10
RP-NAA Reliability Erediction of space vehicle by -
-Monte NAA/NASA 7-11
Carlo simulation
SFRS-W -Simulation of Failure-Responsive Systems -
Westinghouse/NASA 7-12
Rbl4-SBC -Reliability program; computer success probability
-
Service Bureau gorp. 7-13
several components; different distributions; in-
cludes correlation between lifetimes
R1116-SBC Reliability program; computer system reliability -
Service Bureau
corp. 7-13
estimates of components
MARS EP -Mathematical Automated Reliability and SafetyMathematicaISandia 7-14
-Evaluation Program
Equivalently the unreliability is expressed as

1 - R = P{fl *
-
T~ ... TI I= P i a l l t i e sets are bad) (7-3)

or
1- R = PIE1 + c2 ... + E J I=
P t a t least one cut s e t is bad}.
(7-4)

Theabove are e x a c t f o r m u l a s f o rt h es y s t e mr e l i a b i l i t ya n du n r e l i a b i l i t y . Bounds

can be obtained by u s i n g t h e b a s i c p r o b a b i l i s t i c i n e q u a l i t i e s g i v e n below.

R = PIT1 + ... + TI)

T2 + 5 E PITi), (7-5)

R = PITl + T2 + ... + TI) 2 E PETi) - E P{T. T . },etc. (7-6)

i ci I1 I 2
1 2

Thus an upperand a bound %1 lower bound R,l t o . t h e r e l i a b i l i t y are r e s p e c t i v e l y

E$,l = E P{Til (7-7)

I
i
L l = E PITi) - E PITi Ti 1 . (7-8)
i <i 1 2
1 2

In the same manner another upper bound i s o b t a i n e d ,

= C PITi } - E PIT. T . 1 +
1 i <i I1 I 2
1 2 I L 3

The summations are o v e r a l l p o s s i b l e c o m b i n a t i o n o f thesubscriptstaken 2 at-a-time,

3 at-a-time, etc.
S i m i l a r l yt h ei n e q u a l i t i e s (7-5) and (7-6) can b e a p p l i e d t o t h e c u t - s e t form
of t h e e q u a t i o n f o r u n r e l i a b i l i t y (7-4) t oo b t a i n

1- R 2 EPt-6 1
j

or

R 2 1 - EPEE 1 = R , 2 (7-10)
j

and by u s i n g two terms

R 2 1 - EPtc.1
J
+ E P{c. C . 1
= Rv3
(7-11)
j l<j J1 J2

Example: C o n s i d e rt h er e l i a b i l i t yg r a p hg i v e ni n Fig.7-3. Assume independencebetween

items and l e t t h e p r o b a b i l i t i e s o f s u c c e s s f o r e a c h of t h e items be p1 = 0.93,p2 =
0.86, p3 = 0.92,p4 = 0.95, p5 = 0.98.The probabilitiesforthe t i e s and c u t s are
as f o l l o w s :

73
 PIT1) = P(2 5 ) = 0.8428

and

1 - 0.00056
-

Pic2) = 1 - P(2 3 41 = = 0.99944

Upperand lowerbounds forthereliabilityaregiven by u s i n g Eqs. (7-7), (7-8),

(7-9), (7-lo),and (7-ll), respectively,

Rv
l = P{Ti) > 1 ( n ou ts e f ua ls 2 1.)

E$l = 0.843 + 0.838 + 0.866 - P I 1 23 5) - P ( l 24 5) - P(1 3 4 5 )

= 0.2848

%2 = 0.2848 + 0.6850 = 0.9698 = R ( T h i sr e s u l ts h o u l db ee q u a lt o

the system reliability)

%2 = 1 - P{E.I = 1 - 0.03036
0.96964
J
Rv3 = 1 - 03036 + 0.00024 = 0.96988.

As s t a t e d by Messinger[Ref.7-15]theboundsbased on t h e c u t s sets a r e b e s t i n
the high reliability region and thosebased on t h e t i e s e t s are b e s t i n t h e low
reliabilityregion. Hence t h e bounds 1$2 and RV3 a r e t h e p r e f e r r e d bounds i n t h e
aboveexample and.%inthiscasesaves nocomputation as i t i s t h e e x a c t p r o b a b i l i t y
2
of s y s t e m s u c c e s s , as t h e r e a r e o n l y t h r e e t i e sets and t h e bound u s e s a l l c o m b i n a t i o n s
of t i e sets up t o and i n c l u d i n g t h r e e sets.
I n more g e n e r a l p r o b l e m s i n w h i c h t h e r e are J c u t sets t h e number of terms t o
beobtainedinthelower andupperboundscomputations a r e J andJ(J-1)/2respectively.
J
T h i s i s compared t o 2 -1 terms o b t a i n e d by e x p a n d i n ge i t h e r Eq. (7-1) o r (7-4) u s i n g
t i e sets o r c u t sets r e s p e c t i v e l y .
Program D e s c r i p t i o n
Thebounds forsystemreliability,previouslydiscussed,areobtained from cal-
c u l a t i o n sw h i c h are b a s e do nc u t sets. T h i sp r o g r a mc a l c u l a t e su p p e r andlowerbounds

74
u s i n gt h ep r o b a b i l i t i e so fs u c c e s s of each item i n t h e s y s t e m . The program is w r i t t e n
i n FORTRAN. A flowdiagram is giveninFig. 7-4;aprogram l i s t i n g is i n Appendix B.
Inputsimplicity is o n e o f t h e f e a t u r e s of t h i s program.Theuserneedonly
supply the success probabilities and a p r e c e d e n c e list f o r e a c h item i n t h e s y s t e m .
The precedence i s e s t a b l i s h e d by f e e d i n g t o t h e c o m p u t e r v i a c a r d s a l i s t of items
r e s p o n s i b l et ot h ei - t h item. Table 7-2 shows an e x a m p l ec o r r e s p o n d i n gt ot h e relia-
bilitylogicdiagraminFig. 7-3.
The a l g o r i t h m is notcomplex,but is r a t h e r a series o fs i m p l es t e p s .T h e s e
s t e p si no r d e r are: r e a dt h ep r e c e d e n c e l i s t , d e v e l o pt h e t i e sets, d e v e l o pt h e
c u t sets, and c a l c u l a t e t h e bounds.
The precedence l i s t i s c o n v e r t e d t o t h e s u c c e s s p a t h s o r t i e sets by a s u b r o u t i n e
called PATH.
* The arguments are: N, number of items i nt h es y s t e m ; NP, number of
s u c c e s sp a t h sf o u n d ; IP, t h ea r r a yo ft h es u c c e s sp a t h s . The precedence l i s t i s r e a d
by t h e PATH s u b r o u t i n e ; i t s format i s d i s c u s s e du n d e rt h ei n p u td e s c r i p t i o n .A f t e r
beingprintedthepaths a r e c o n v e r t e dt o a Boolean a r r a y of z e r o s and o n e s , a n d t h e
c u t sets are developed by theproceduregivenbelow. When t h e c u t sets are a v a i l a b l e
t h e bounds are c a l c u l a t e d by a p r o c e d u r e in Ref. 7-15.
Generationof Cut S e t s
A simple procedure using Boolean logic is u s e d f o r o b t a i n i n g a matrix identifying
t h em i n i m a lc u t s of t h es y s t e mf r o mo n ec o n t a i n i n gt h ep a t h s . L e t thepathmatrixbe

T a b l e 7-2
Precedence L i s t f o r Program Input

ITEM PREDECESSORS CARD CODE

1 IN -1

2 IN -1

3 1 1

4 1 1

5 2, 3, 4 2 , 3, 4

OUT 5 20

*T h i s a l g o r i t h m w a s obtained from Naval Applied Science Labs ofBrooklyn, N. Y.

75
Read t h e number of elements and the pro-
abilitv of success of each

list foreachelement
anddeterminethe
p a t h s i n an a r r a y
c a l l e d IP

I
Do 'a m a t r i x
"multiplication"
usingthelogical - -- - - Return
"OR" s t a t e m e n t t o
c a l c u l a t e IP?

Comment: The p a t h sa r ed e t e r m i n e d
Determinethe s i m p l e by element number i n r e v e r s e o r d e r ;
anddoubleelement t h e r e f o r e , t h e program c o r r e c t s t h e
c u t s by l o o k i n g a t
o r d e rf o ro u t p u tp u r p o s e s andforms
IPP array a Boolean matrix whereby the paths
a r e t h e rows.

Determine t r i p l e
e l e m e n tc u t s by
p e r f o r m i n gt h e
"OR" o p e r a t i o n
on a l l p o s s i b l e
t r i p l e products

starting with lower

F i g u r e 7-4. Flow Diagram f o r ComputerProgram--Bounds f o rR e l i a b i l i t y .

76
 1 0 1 0 1

P = 0 1 0 0 1

1 0 0 1 1

where t h ep a t h s are p1 = 1, 3 , 5 ; p2 = 2 , 5 ; and p = 1, 4 , 5 , respectively. Now

3
c o n s i d e rt h e column v e c t o r s 1, 0, 1; 0, 1, 0; etc. of t h ep a t hm a t r i x P. For a
singleelementtobe a c u t , i t mustbe i n e a c hp a t h ; i. e., i t s column v e c t o r i n P
m u s tb et h eu n i tv e c t o r (1, 1, 1). Note t h a te l e m e n t 5 i s theonlyelementwhich is
c o n t a i n e di n a l l p a t h s ;h e n c e 5 is t h eo n l ys i n g l ee l e m e n tc u t .I ng e n e r a l ,i f PC
d e n o t e s a column v e c t o r of ann-path matrix, t h e n f o r

t h ec o r r e s p o n d i n ge l e m e n t c i s a s i n g l ee l e m e n tc u t .I f P = 0 f o r some i i n each
paththenthere are no s i n g l e e l e m e n t c u t s andonemustproceed tolookfor two
elementcuts.
For two e l e m e n t c u t s c o n s i d e r f o r c # d

where the "+" i n d i c a t e st h el o g i c sum o ru n i o n .I f

PC + Pd = 1, f o r a l l i=1,2,...,n,
i i

thenelements c and d form a two e l e m e n t c u t .

Thisprocedurecontinuesuntil a l l possiblecutsoforder 1, 2 , . . . , n havebeen
exhaustedoruntilonlyunitvectors are o b t a i n e d i n t h e v e c t o r u n i o n s as d e s c r i b e d .
A t eachstage a l l t h e nonminimal c u t s are e l i m i n a t e d by u s i n g t h e f o l l o w i n g a p p r o a c h .
After a p o s s i b l e c u t of o r d e r M h a s b e e n i d e n t i f i e d , i t i s checkedagainst all cuts
o fo r d e r M-1, M-Z,..., 1 by u s i n g B o o l e a n l o g i c f o r i n t e r s e c t i o n , i.e., t h e AND
operation,forthemultiplicationof two v e c t o r s ; i f t h e p o s s i b l e c u t c o n t a i n s a
c u t of smaller o r d e r t h e v e c t o r p r o d u c t w o u l d b e e q u a l t o t h e o r d e r of t h e smaller
cut. A l l c u t s a r e e l i m i n a t e df o r which t h i s v e c t o r p r o d u c t as d e f i n e d i s e q u a l t o
theorder of t h e s m a l l e r c u t .
Theabove stepsdescribe how t h e program i d e n t i f i e s minimalcuts,even to the
"OR" l o g i c u s e d t o f o r m t h e v e c t o r u n i o n .

77
Input and Program Limitations
There are t h r e e b a s i c i n p u t s t o t h e program.The i n p u tv a r i a b l e s are N, t h e
number o f items i n t h e s y s t e m , PROB, t h e p r o b a b i l i t i e s o f s u c c e s s of e a c h item,
IACTIV, t h e i - t h item, and IPRED, t h ei t e m ( s )i m m e d i a t e l yp r e c e d i n gt h e IACTIV item.
The l i m i t on t h e number, N , of items is 20 n o t i n c l u d i n g t h e e n d p o i n t s ; f o r N
t h ef o r m a t is (15).There i s no limit on PROB; however,each item shouldhave
s p e c i f i e d a p r o b a b i l i t yo fs u c c e s s ;f o r m a t i s (8E10.4).
IACTIV and IPRED are v a r i a b l e sa s s o c i a t e dw i t ht h ep r e c e d e n c e list. IACTIV
is t h e item a c t i v e l y u n d e r c o n s i d e r a t i o n , and IPRED i s a v e c t o r of items t h a t p r e c e d e
t h e item IACTIV. Ifthe item I A C T I V is preceded by t h e i n p u t p o i n t IPRED i s t h e s i n g l e
number -1, and i f succeeded by t h eo u t p u tp o i n t i t is t h e number20. I A C T I V may be
any number up through20;the IPRED v e c t o r may have a t most 9 numbers.There will
be N + l i n p u t c a r d s , o n e f o r e a c h e l e m e n t i n t h e l o g i c m o d e l andone for the output
n o d e ;t h ei n p u te l e m e n tf o r m a t i s (1015).
output
The o u t p u t i s b r i e f and e a s i l y r e a d . I n p u t p r o b a b i l i t i e s are p r i n t e d and followed
by t h e t i e s e t s and c u t sets.
Sincethecalculationfor bounds is done by adding terms t o a series witheach
new term r e s u l t i n g i n a new bound, e i t h e r l o w e r o r u p p e r ,t h e bounds a r e g i v e n a t
e a c hs t e pw i t ht h ea p p r o p r i a t e l a s t term shown. For small systemstheexactsystem
r e l i a b i l i t y is c a l c u l a t e db e f o r et h ep r o g r a m i s terminated.
Example:The example i n F i g . 7-3 is used.
The p a t h m a t r i x i s g i v e n by

1 2 3 4 5 Paths
"-"""""
1 0 1 0 1 1,3,5

P = 0 1 0 0 1 2Y5

1 0 0 1 1 1,4,5

and t h e c u t m a t r i x by

1 2 3 4 5 cuts

0 0 0 0 1 5

c = 1 1 0 0 0 1,2

0 1 1 1 0 2,3,4
T h et h r e ec u t s are t h u s 5;l and 2; and 2 , 3 , and 4. Theupperandlowerbounds are
o b t a i n e d as i n d i c a t e di nt h ep r e v i o u sd i s c u s s i o n . The program r e s u l t s as shown i n
T a b l e 7-3. h a v e b e e n r e t y p e d f r o m t h e c o m p u t e r p r i n t o u t .

T a b l e 7-3

Bounds f o r System R e l i a b i l i t y Example

CIRCUIT CONTAINS 5 ELENENTS

ELEMENT PROBABILITY
NUMBER OF SUCCESS

.9300
.8600
.9200
.9500
.9800

TIE SETS OR SUCCESS

PATHS ( 3)

ELEMENT
PATH NUMBER5

2 5
1 3 5
1 4 5

CUT SETS ( 3)

1 5
2 1 2
3 2 3 4

LOWER BOUND I S .96964E 0 LAST TERM -30361E -1

UPPER BOUND .969883

IS 0 LAST TERM .24641E -3

LOWER BOUND I S .96988E 0 LAST TERM .78407E -6

SYSTEM RELIABILITY .98988E 0

79
 F i g u r e 7-5. SystemDiagram f o r Bounds Program Example 2
w
Example 2
The system shown i n F i g . 7-5 i s used; i t i s a r e l a t i v e l y complex s e r i e s - p a r a l l e l
network. The i n p u tt ot h es y s t e m is a t t h el e f t of t h ef i g u r e . As canbeseenfrom
thefigure,there a r e many p o s s i b l es u c c e s sp a t h st h r o u g ht h es y s t e m , andhand calcu-
lationofsystemreliability would be a t b e s t v e r y t e d i o u s . The r e l i a b i l i t y of each
element is g i v e ni nT a b l e 7-4. A s r e q u i r e d by t h ep r o g r a m ,e l e m e n tf a i l u r e s are
assumed independent. The boundsprogram p r i n t o u tf o l l o w s . A s c a nb es e e n from t h e
last two l i n e s of t h e p r i n t o u t , t h e programhas bounded t h es y s t e mr e l i a b i l i t y .S i n c e
theupperandlowerboundshaveconvergedtothe same v a l u e ,0 . 9 7 7 2 6 ,t h i sv a l u e is
thesystemreliabilityto5-placeaccuracy.

T a b l e 7-4
R e l i a b i l i t i e s ofElements i nF i g . 7-5
E l . No. 1 2 3 4 5 6 7 8 9 10 111 3 1 2 1 4 16 15
Rel. .80 .80 .90
.85
.75
.87
-82
.82
.89 .88 .85 .85 .85
.75
.70

80
 Table 7-5
Program P r i n t o u t f o r Example 2

B Q U N ' . J S F U N S Y S T E M K ~ L I A B I L I I Y
C I H C U I T C O N T A I N > 16 ELEHtNTS

PHOBA&ILITY
(iF SUCCESS

1 .8000
2 .moo
3 .9oao
4 ,0500
5 .7500
6 .8700
i ,8200
8 .8200
9 8900
10 ,8800
11 ,8500
12 ,8500
13 .8501r
14 .7bOU
15 .7LiOb
16 .7000

1 1 4 7 9 11
2 1 4 7 9 12
3 1 4 7 9 13
4 1 4 14 15
5 1 4 14 16
0 1 4 10 11
7 1 4 10 12
8 1 4 10 13
9 1 4 8 9 11
Table 7-5 (Cont d)
10 1 4 R Y 12
11 1 4 8 9 13
12 1 5 14 15
13 3 6 14 15 I-.

14 1 5 14 16
15 3 6 14 16
16 1 5 10 11
17 3 6 10. 11

3 6 10 12
1 5 10 13
3 6 10 13
1 5 7 9 11
3 6 7 9 11
1 5 7 9 12
3 6 7 9 12
1 5 7 9 13
3 6 7 Y 13
2 4 14 15
2 4 14 16
2 4 10 11
2 4 10 12
2 4 10 13
1 5 8 9 11
3 6 8 9 11
1 5 8 Y 12
3 6 0 9 12
1 5 8 9 13
3 4 0 9 13
2 5 14 15
2 5 14 16
2 5 10 11
2 5 10 12
2 5 10 13
2 4 7 9 11
2 4 7 9 12
2 4 7 9 13
2 4 8 9 11
2 4 8 Y 12
2 4 8 Y 13
 Table 7-5 (Continued)
50 2 5 7 9 11
51 2 5 7 9 12
52 2 5 7 9 13
53 2 5 8 9 11
5s 2 5 8 9 12
55 2 5 8 9 13

CUT skTS( 1u
1 1 2 3
2 1 2 6
3 3 4 5
4 4 5 6
5 9 10 14
6 7 8 10 14
7 9 10 15 16
8 11 12 13 14
9 7 8 10 15 16

I
 7.3.2. R e l i a b i l i t y C o s t Trade-Of f Analysis Program
The % l i a b i l i t y C o s t x r a d e - o f f A n a l y s i s (RECTA) p r o g r a m o b t a i n s an optimumcon-
figurationfor a s y s t e n ic o n t a i n i n gs p a r e , active andstandbycomponents. The system
configurationinitiallycontains no r e d u n d a n c y i n v o l v i n g i d e n t i c a l e l e m e n t s , b u t may
haveredundant elements with d i f f e r e n t f a i l u r e - r a t e c h a r a c t e r i s t i c s . T h i s program
combines some of the f e a t u r e s o f t h o s e d e s c r i b e d i n Refs. 7-16 and 7-17. Theprogram
is l i s t e d i n Appendix C.
The main f e a t u r e o f t h e program is a s u b r o u t i n e w h i c h c a l c u l a t e s t h e r e l i a b i l i t y
of a n e l e m e n t c o n t a i n i n g :
(1) n i d e n t i c a la c t i v ep a r a l l e l items, a t l e a s t n ofwhichmustoperate,
0
(2) m i d e n t i c a l spares, and
(3) r i d e n t i c a ls t a n d b yr e d u n d a n t items.
The c o m p u t a t i o n a s s u m e s i n d e p e n d e n c e a n d t h e e x p o n e n t i a l f a i l u r e time d i s t r i b u t i o n .
Volume I V - Prediction of this series c o n t a i n s a c o m p l e t e d e s c r i p t i o n of t h e p r o c e d u r e .
The s y s t e m r e l i a b i l i t y model g i v e s t h e p r o b a b i l i t y of s u c c e s s f u l o p e r a t i o n o f t h e
system i n terms o ft h ee l e m e n tr e l i a b i l i t i e s . The s y s t e m r e l i a b i l i t y is calculated
by a model s u p p l i e d i n a s u b r o u t i n e by t h e u s e r . The u s e r a l s o s u p p l i e s i n d i c a t o r s
f o re a c he l e m e n tf o rt h et y p e s of redundancyhewishes t o c o n s i d e r ; a o n e( 1 )i n d i c a t e s
thattheparticular, formofredundancy i s p e r m i t t e d ,a n d a z e r o (0) i n d i c a t e s t h a t no
items o f t h e p a r t i c u l a r r e d u n d a n c y t y p e may be added.
One a d d i t i o n a l f e a t u r e o f t h e program is t h e h a n d l i n g o f m a j o r i t y v o t i n g l o g i c .
An upper limit i s s u p p l i e d a s t h e i n d i c a t o r i n p u t v a l u e . The items w i l l b ei n c r e -
mented i n s t e p s of 1, 3 , 5, . . ., N where N i s t h e limit provided by t h e u s e r . An ,

exampleof majorityvoting i s f o r 5 items i n an element, a t l e a s t 3 of whichmust

work.
Startingwiththeinitialsystemconfiguration, a l l possiblesingle item addi-
t i o n s (two items i n t h e case of m a j o r i t y v o t i n g e l e m e n t s ) are made and t h e i n c r e a s e
in the system reliability is o b t a i n e d f o r e a c h c o n f i g u r a t i o n by t h e e l e m e n t relia-
bilitysubroutine and t h e s u b r o u t i n e s u p p l i e d by t h e u s e r f o r t h e c o m p u t a t i o n of t h e
r e l i a b i l i t y o f t h es y s t e m . The i n c r e a s e i n c o s t is a l s o computed f o re a c hc o n f i g u -
r a t i o nu s i n gt h ei n p u tc o s ti n f o r m a t i o n . The r a t i o s o f t h e i n c r e a s e i n r e l i a b i l i t y
t o the increaseincost a r e computed f o r e a c h p o s s i b l e a l t e r n a t i v e a s s p e c i f i e d by
t h ei n d i c a t o r s . The redundant item y i e l d i n g t h e g r e a t e s t r a t i o i s t h eo n es e l e c t e d
foradditionandtheprocedure is repeated for the next step starting with the new
configuration.
* Theprogram c o n t i n u e su n t i l a c o n v e r g e n c ec r i t e r i o n ,s u p p l i e d by

* T h i sa l g o r i t h my i e l d sa ni n c o m p l e t eu n d o m i n a t e ds e q u e n c e of o p t i m a l s o l u t i o n s i n
t h e case of a s e r i a l s y s t e m i n i t i a l l y . I n t h e case of n o n s e r i a ls y s t e m st h ep r o c e d u r e
may not y i e l d a n o p t i m a l s e q u e n c e of s o l u t i o n s a l t h o u g h i t would b e e x p e c t e d t o y i e l d
n e a ro p t i m a lc o n f i g u r a t i o n s .S e eR e f .7 - 1 8 c o n c e r n i n gt h i sp o i n tf o r s e r i a l systems.
t h eu s e r ,h a sb e e ns a t i s t i f i e d .F o re x a m p l e , i t may c o n t i n u e u n t i l t h e i n c r e a s e i n
r e l i a b i l i t y i s less t h a n 0.001.
By v i r t u e of t h e i n d i c a t o r s a featureofthis programwhich is n o t o b v i o u s is"
t h a t i t canbeusedfor a spares allocation procedure based on either one of two
criteria:
(1) m i n i m i z es t o c k o u tp r o b a b i l i t ys u b j e c tt o a g i v e nc o s t ,o r
(2) maximizesystem reliabilitysubjectto a g i v e nc o s t .
In the l a t t e r c a s et h es y s t e mc o n f i g u r a t i o n is used i n t h e r e l i a b i l i t y c o m p u t a t i o n
whereas i n t h e f o r m e r t h e e l e m e n t s are c o n s i d e r e d t o b e i n series.
I n p u tD e s c r i p t i o n
The i n p u t i s s t r a i g h t f o r w a r dw i t ho n eo p t i o n a li n p u t . A b r i e fe x p l a n a t i o n is
g i v e nf o re a c hi n p u tc a r da n d i t s variables;these are followed by anexample.
The f i r s t two c a r d s i d e n t i f y t h e s y s t e m b e i n g a n a l y z e d w i t h t h e f i r s t c a r d
having two systemparameters NEL and CONVG. NEL i s t h e number ofelements inthe
systemand CONVG i s t h es y s t e mr e l i a b i l i t yc o n v e r g e n c ec r i t e r i o n . When t h e i n c r e a s e
i n r e l i a b i l i t y i s less t h a n CONVG t h e programbranchestoread new d a t a . The second
cardhasanidentificationfortheproblembeingrun;all 80 columns may beusedand
themessage is notrestrictedastotype of c h a r a c t e r s .
The i n f o r m a t i o nf o re a c he l e m e n t is nextreadintheorderspecifiedinthe
systemmodel. The elementparameters a r e d e f i n e di nt h ef o l l o w i n gt a b l e .

T a b l e 7-6
I n p u t Card V a r i a b l e Names
Variable Identification
Card 1 TIME Length of mission
FRATE Failure rate
RELSW Switch r e l i a b i l i t y
ELCST A c t i v e item c o s t
SPCST Spare cost
SWCST Switchcost
RSCST Redundantstandbycost
NO Minimum number of items n e c e s s a r y
foroperation
Card 2 N
ID I n d i c a t o r s oftypeofredundancy
permitted
INPRM I n i t i a l number of items i n t h e
syst e m .

85
 The o p t i o n m e n t i o n e d a b o v e c o n c e r n s t h e v a r i a b l e N
ID which may i n d i c a t e m a j o r i t y
v o t i n g .I ft h i s is d e s i r e da nu p p e r limit i s i n s e r t e d as t h ei n d i c a t o r . The program
w i l l eliminate the particular variable from consideration when i t h a s b u i l t up t o t h e
specified l i m i t . The m a j o r i t y v o t i n g a p p l i e s t o a c t i v e items o n l y .

Table 7-7
Example of InputCards
9.1000E-02
,MAJORITY VOTING LOGIC WITH REDUNDANT STANDBYS I N THE LAST TWO ELEMENTS
.1000E 03
-51303-03
.9900E 0 0 . .2000E 0 1 .2000E 01 .2000E 00 .2000E 0 1 1
3 0 0 1 0 0
.1000E 03 .5130E-03
.9900E 00 .2000E 0 1 .2000E 01 .2000E 00 .2000E 01 1
3 0 0 1 0 0
.1000E
03
.5130E-03
.9900E 00 .2000E 01 .2000E 01 .2000E 00 .2000E 01 1
3 0 0 1 0 0
.1000E
03
.5130E-03
.9900E 00 .2000E 0 1 .2000E 01 .2000E 00 .2000E 01 1
3 0 0 1 0 0
.1054E-02 .1000E 03 .1000E 02 1
3 0 0 1 0 0
.
.4000E .1054E-02
lOOOE 03 01 1
3 0 0 1 0 0
00E .1054E-02 .1000E 03 01 1
3 0 0 1 0 0
.1000E 03
.6931E-02
.9900E 00 .1000E 03
.1000E 03 .1000E 02
.1000E
03 1
3 0 1 1 0 0
.1000E
03
.2877E-02
.9900E 00 .30QCIE 02
.3000E
02
.3000E 01 .3000E
02 1
3 0 1 1 0 0

OutputDescription
Initial values of t h e p a r a m e t e r s and o t h e r p e r t i n e n t i n f o r m a t i o n a b o u t t h e s y s t e m
c o s t are p r i n t e d f i r s t f o r i d e n t i f i c a t i o n . The i n i t i a l r e l i a b i l i t y is calculated
and p r i n t e df o re a c he l e m e n ts e p a r a t e l y .T h i s i s followed by a summary of theelement
i n f o r m a t i o n and t h e s y s t e m r e l i a b i l i t y and c o s t f o r a systemconsisting ofno
redundancy.
The i t e r a t i o n b e g i n s by p r i n t i n g t h e e l e m e n t r e l i a b i l i t y w i t h o n e i t e m a d d e d
wheredesignated by i n d i c a t o r s . One of t h e s ea d d i t i o n s( s p a r e ,s t a n d b y ,o ra c t i v e
parallel) is selectedforthe optimum c o n f i g u r a t i o n w i t h r e s p e c t t o r e l i a b i l i t y and
cost. The r e l i a b i l i t y of t h i s same element i s c a l c u l a t e dw i t ho n ea d d i t i o n a l i t e m of
redundancyofeachtypepermitted. The r a t i o s of i n c r e a s e i n r e l i a b i l i t y t o i n c r e a s e

86
o fc o s t are compared f o r t h i s e l e m e n t and o t h e r s c a l c u l a t e d earlier f o r t h e optimum
configuration a t t h i ss t a g e . When t h e optimum is f o u n d t h e i n f o r m a t i o n f o r t h i s
s t e p i s p r i n t e d and t h e programproceeds t o t h e next s t e p . Theprogram repeatsthe
aboveprocedureaddingoneredundant item a t a time t o a s e l e c t e d e l e m e n t u n t i l t h e
convergencerequirement is m e t .
3 major steps in RECTA are summarizedbelow.
I n i t i a ls t e p :d a t a is r e a d a n d t h e r e l i a b i l i t y is c a l c u l a t e df o re a c h
element a t i t s i n i t i a l state. The i n i t i a l s y s t e m r e l i a b i l i t y and c o s t are
alsocalculated.
I n t e r m e d i a t es t e p :e a c h item of eachelementthat i s allowed t o v a r y is
i n c r e m e n t e ds e p a r a t e l ya n dt h ei n c r e a s e si nr e l i a b i l i t ya n dc o s t of t h e
system a r e c a l c u l a t e d .
I t e r a t i o nl o o p :t h el o o pb e g i n s by c h o o s i n gt h ec o n f i g u r a t i o ng e n e r a t e d
intheintermediatestepthatyieldsthebestcost-reliabilitytrade-off.
The item t h a t i s added t ot h es y s t e m is t h e n r e p l a c e d i n t h e i n t e r m e d i a t e
s t a t e by i t s n e x ti n c r e m e n t ;t h u s ,t h ei n t e r m e d i a t es t a t ea l w a y s i s one
s t e p ahead of t h e s y s t e m c o n f i g u r a t i o n .
p r o g r a mc o n t i n u e st oq u e r yt h ei n t e r m e d i a t ev a l u e s andaddcomponents until
thesystemreliabilitysatisfiestheconvergencecriterion.
Example
Thisexample is a s i m p l i f i e d b l o c k d i a g r a m of a c o m p u t e r c o n t a i n i n g n i n e (9)
b l o c k s( e l e m e n t s ) assumed t o b e i n series l o g i c , as shown i nF i g . 7-6. A l l elements

F i g u r e 7-6. S i m p l i f i e d Computer Block Diagram f o r RECTA Example

87
 Table 7-8
RECTA Program Example

o99000
.9YOOO
-99090
2.00
2 .OU
2.0u
2.00
2.00
2.00
.
2.00
2 0.u
2.00
20
20
20
e99000 2.00 2.00 2. ou ao
0.00000 10.09 0;oo 0 .ou O*QO
0 .ouooo 4 *OU 0.00 0 000 0 *a0
0 .ooooo 4.00 0 :on 0 .ou 0.00
03
e99000 lUO*OU 1oo;oo 1oo.ou 10 .PO
03 .99000 JO.0U 30 0.00 30.00 3.p0
Table 7-8 (Cont 'd)
"""_""""~""""""

,949YYJ
,949Y9.5
,949993

..
949Y93
a99963
BY9963
.8Y9Y63
,500023
749Y87
Table 7-8 (Cont'd)
""""""""-"""""""~"""""""""""

KkLIARILlTY
tS'fIHATES FOH E L t H t N T 9 CUNlAININb

STAND~Y RtL,
U ,843731E 0

1
1
0
1
.YYOO
OU29
100.0

STAIQDOY RtL.
II e749987E 0
1 ,965755E U
Table 7-8 (Cont'd)

d 0 0 .992/44
1 Li 0
1 0 0
1 u 0
A 0 0
1 0 0
1 U 0
1 U 0
1 0 0

SYSTEM RtLlAoICIry ,232655

"""" s
y
sr
EH
-
~~
'"
"
"_
LS
I
Lc
Pe
"
""
-"
"
""
""
"
""
""
can be.made redundant by u s i n g m a j o r i t y v o t i n g l o g i c a n d two e l e m e n t s c a n b e f u r t h e r
modified by u s i n gs t a n d b ye l e m e n t sw i t hs w i t c h i n g . The i n p u ti n f o r m a t i o n is contained
o nt h ef o l l o w i n gp r i n t o u t . The i n i t i a l s y s t e m c o n f i g u r a t i o n i n d i c a t e s t h e number of
active, s p a r e , a n d s t a n d b y e l e m e n t s i n t h e s y s t e m a t t h eb e g i n n i n go ft h ec o m p u t e r
run,theindicator t e l l s t h e computer t h e e l e m e n t s w h i c h c a n b e made redundant by
addingfurther a c t i v e items o r u s i n g m a j o r i t y v o t i n g l o g i c , a d d i n g s p a r e s a n d
s t a n d b y s .I nt h i s example t h e m a j o r i t y v o t i n g l o g i c i s used(a maximum of t h r e e
elements a t l e a s t two ofwhichmustoperate),nospares a r e permitted,andstandby
items are p e r m i t t e df o re l e m e n t s 8 and 9. The programoutput is g i v e n i n T a b l e 7-8.
The program o b t a i n s t h e r e l i a b i l i t y estimates f o r e a c h e l e m e n t s u b j e c t t o
i t s i n i t i a lc o n f i g u r a t i o n ;t h a tf o re l e m e n t 1 i s shown below. Then t h e r e l i a b i l i t y
o ft h ei n i t i a ls y s t e m is computedfrom t h e model s u p p l i e d by t h eu s e r . The i n i t i a l
s t e p i s g i v e ni nt h ep r i n t o u t . A t t h i sp o i n tt h e program is r e a d yt o alter
eachelement i n a l l p o s s i b l e manners as s p e c i f i e d by t h e i n d i c a t o r s i n o r d e r t o
determinethe optimum c o n f i g u r a t i o n f o r o n e item added(two forthemajorityvoting
alternative). The r e s u l t s f o r item 9 a r e g i v e nb e c a u s et h e r e are two a l t e r n a t i v e s .
The i t e m w h i c h g i v e s t h e l a r g e s t i n c r e a s e i n r e l i a b i l i t y p e r u n i t c o s t i s t h eo n e
selectedforstep 1; i n t h i s c a s e i t i s element 1 and a m a j o r i t yv o t i n ge l e m e n t
w i t ht h r e e items i s used.Thisprocedure i s r e p e a t e d a t e a c hs t e pt oo b t a i n a
system configuration with the desired relaibility or one for which the increase in
r e l i a b i l i t y i s less t h a n 0.001. F o u r t e e ns t e p s were used i n t h e a n a l y s i s ; t h e f i n a l
systemconfiguration is g i v e n on t h e f i n a l p r i n t o u t a l o n g w i t h t h e s y s t e m r e l i a b i l i t y
a n dc o s t .

References

7-1. Van T i j n , D. E . : D e s c r i p t i o no ft h eC o m p u t e r i z e dR e l i a b i l i t yA n a l y s i s Method

(CRAM). ARINC Research Monograph 11, 1964.

7-2. Whiteman, I. R.: RESCRIPT--A ComputerProgrammingLanguage f o rR e l i a b i l i t y ,

P r e s e n t e d a t F i f t h Annual West C o a s t R e l i a b i l i t y Symposium, Los Angeles,
C a l i f o r n i a ,1 9 6 4 .

7-3. House , J. F. ; a n dL a C a p r a ,J o h n :S y s t e m sR e l . i a b i l i t yA n a l y s i sa n dP r e d i c t i o n
t h r o u g ht h eA p p l i c a t i o no f a D i g i t a l Computer.Presented a t National
Symposium onSpaceElectronicsandTelemetry, M i a m i Beach,Florida,1962.

7-4. S h e l l e y , B. F . ; andHamilton, DC. . : A Mechanized A i r c r a f tR e l i a b i l i t yA n a l y s i s

Model. P r o c e e d i n g so ft h eT e n t h Symposium on R e l i a b i l i t y a n dQ u a l i t yC o n t r o l ,
Washington, D. C., 1964.

7-5. McFaul, C h a r l o t t e : Deep SubmergenceRescue Vessel R e l i a b i l i t yP r e d i c t i o n .

T e c h n i c a l Memo 415165. US Navy MEL, Annapolis,Maryland,1965.
References(Continued)

7-6. Weisberg, S . A. ; andSchmidt, J . H. : Computer Techniques f o rE s t i m a t i n gS y s t e m

R e l i a b i l i t y .P r o c e e d i n g so ft h e1 9 6 6A n n u a l Symposium on R e l i a b i l i t y , San
F r a n c i s c o ,C a l i f o r n i a ,1 9 6 6 , pp. 87-97.

7-7. Kiefer, F. P.; e t . a l . : Man-rating t h e GeminiLaunch V e h i c l e (Crew Hazard

andMissionAnalysis).Proceedings of t h e 1966Annual Symposium on
R e l i a b i l i t y , San F r a n c i s c o ,C a l i f o r n i a ,1 9 6 6 ,p p . 87-97.

7-8. Finch, R. E.: An SPS S u b r o u t i n e as a SimulationAid. Master o fS c i e n c eT h e s i s .

S c h o o lo fE n g i n e e r i n g , A i r U n i v e r s i t y ,W r i g h t - P a t t e r s o n AFB, AD 425237,1963.

7-9. O t t l i n g e r , J . A.; e t . a l . : SurveyofStudiesand ComputerProgramming Efforts

f o rR e l i a b i l i t y ,M a i n t a i n a b i l i t ya n dS y s t e mE f f e c t i v e n e s s .R e p o r t OEM-1,
OfficeoftheDirectorofDefenseResearchandEngineering,Department of
Defense, AD 622 676.

7-10. McKnight, C. W . ; e t . a l . : An A u t o m a t i cR e l i a b i l i t yM a t h e m a t i c a l Model.

ProceedingsoftheEleventhNational Symposium on R e l i a b i l i t y a n d Q u a l i t y
C o n t r o l , Miami Beach,Florida,1965.

7-11. Hershkowitz, B. H . ; e t . a l . : R e l i a b i l i t yS i m u l a t i o n Model. P r o c e e d i n g so ft h e

Tenth National Symposium on R e l i a b i l i t y and Quality Control, Washington,
D. C . , 1964.

7-12. Hannigan, J . M.: A Computer Program f o r t h e S i m u l a t i o n of Failure-Responsive

Systems.TechnicalReport No. 6,WestinghouseDefenseandSpaceCenter,
STAR N66-26880, 1966.

7-13. House, J . F.;andLaCapra,John: R e l i a b i l i t yE n g i n e . e r i n g a t SBC. Brochure

fromtheServiceBureauCorporation/ComputingSciencesDivision,1966.

7-14. Hubbard, R. L.: The MarsepProgram. Two pagedescriptionfromMathematica,

P r i n c e t o n , New J e r s e y .

7-15. Messinger, M.; and Shooman, M. L . : R e l i a b i l i t yA p p r o x i m a t i o n sf o r Complex

S t r u c t u r e s . 1967Annual Symposium on R e l i a b i l i t y , Washington, D. c., January
10-12,pp. 292-301.

7-16. T r o t t , E . P . :M a i n t a i n a b i l i t y and R e l i a b i l i t y CostEffectivenessProgram

(MARCEP). F o u r t hA n n u a lR e l i a b i l i t ya n dM a i n t a i n a b i l i t yC o n f e r e n c e ,
July 1965,pp. 219-228.

7-17. P a r r , V. B.: Automated R e l i a b i l i t yT r a d e - o f f Program-ARTOP 11. P r o c e e d i n g s

1967Annual Symposium on R e l i a b i l i t y , J a n u a r y 10-12, 1967,pp. 847-850.

7-18. Barlow, R. E . ; andProschan,F.:MathematicalTheory of R e l i a b i l i t y . John

Wiley & Sons,Inc. , N e w York,1965,256p.

7-19. S a n d l e r , G. H.: System R e l i a b i l i t yE n g i n e e r i n g P

. r e n t i c e - H a l l I, n c . , Englewood
C l i f f s , N e w J e r s e y ,1 9 6 3 , 221p.

7-20. Meyers, R. H.: R e l i a b i l i t yE n g i n e e r i n gf o rE l e c t r o n i cS y s t e m s .J o h nW i l e y &

Sons,Inc., New York,1964,360p.

93
8. Testing
Many o f t h e r e s u l t s o f e x p e r i m e n t a l p r o g r a m s c a n b e a n a l y z e d b y g r a p h i c a l t e c h n i q u e s
such as drawing a c u r v e by freehand through a set o f d a t a p o i n t s , o r bycomparing a
t e s t ,measurementwith a p h y s i c a lr e q u i r e m e n t .T h e s ep a r t i c u l a rm e t h o d so fa n a l y s e s
do n o tr e q u i r ef o r m a lc o m p u t a t i o n by t h e u s e o f a d i g i t a l computer.However,it is
not unusual in typical experimental programs to encounter situations in which one is
measuringseveralperformanceattributesand as many as 1 0 o r more i n d e p e n d e n t v a r i a b l e s
such as p a r t c h a r a c t e r i s t i c s and -environmental stresses. I no r d e rt oa n a l y z ed a t ao f
thiscomplexity i t i s u s u a l l yn e c e s s a r yt ou s ed i g i t a lc o m p u t e rp r o g r a m sw h i c h are
already available.
Inaddition,one is o f t e n f a c e d w i t h t h e p r o b l e m o f e s t i m a t i n g t h e p a r a m e t e r s
of l i f e d i s t r i b u t i o n s on t h e b a s i s of anobservedsampleof items p l a c e d on t e s t f o r
a f i x e d test time. I no r d e rt oh a v et h ec a p a b i l i t yo fd e s c r i b i n gt h e s ed a t a bymeans
ofone o r more of t h e many f a i l u r e - t i m e d i s t r i b u t i o n s , i t i s c o n v e n i e n tt oh a v e com-
puterprogramstoperformthetediousanalyses.
Inthissectionthecomputationalapproaches are s u b d i v i d e d i n t o t h o s e w h i c h
p e r t a i nt o :( 1 )a t t r i b u t ed a t a ,( 2 )v a r i a b l e sd a t a , and ( 3 ) s t r e s s - s t r e n g t hm e a s u r e -
ments. By a t t r i b u t e d a t a w e mean s i m p l yt h a tt h eo b s e r v a t i o n ofanexperiment is
c l a s s i f i e d as a f a i l u r e o r n o n f a i l u r e , o r i n a case of a performancemeasurement that
t h eo b s e r v a t i o n is c l a s s i f i e d as go o r no-go. I nt h e l a t t e r c a s e ,t h er e g i o n of
o b s e r v a t i o n s is s u b d i v i d e d i n t o two d i s j o i n t r e g i o n s ; t h e a c c e p t a b l e p e r f o r m a n c e
r e g i o na n dt h en o n a c c e p t a b l er e g i o n - . By v a r i a b l e s d a t a w e mean observationswhich
cantakeonanyoneof a s e t o fv a l u e so v e r a g i v e nr a n g eo fv a l u e s . The t h i r d
c a t e g o r y ,s t r e s s - s t r e n g t hm e a s u r e m e n t si n c l u d e ss t r e s s - a t - f a i l u r ed a t a ,s u c h as would
b eo b t a i n e di n a t e n s i l e test of a p a r t i c u l a r metal specimen. I t a l s oi n c l u d e st h ?
dataresultingfromsensitivitytesting,wherean item i s p l a c e d on t e s t a t a f i x e d
stress l e v e l and test r e s u l t s r e c o r d e d as a f a i l u r e o r a n o n f a i 1 u r e . T a b l e 8-1 summarizes
the results of t h i s s e c t i o n w i t h r e s p e c t t o t h e t y p e of d a t a and t h e a s s o c i a t e d
problems.Table 8-2 c o n t a i n s a l i s t i n g of thecomputerprogramswhich may b e h e l p f u l
insolvingthecorrespondingproblems.
8.1 A t t r i b u t e Data
The t y p i c a l c o m p u t a t i o n a l p r o b l e m s a s s o c i a t e d w i t h a t t r i b u t e d a t a are t o p r o v i d e
s a m p l i n gp l a n sa n dt h e i ro p e r a t i n gc h a r a c t e r i s t i c s and t o o b t a i n c o n f i d e n c e limits
f o rt h et r u ep r o p o r t i o no fn o n f a i l u r e s( o r "go" i t e m s ) . Bothof t h e s ep r o b l e m su s u a l l y
are s o l v e du s i n gt h eb i n o m i a ld i s t r i b u t i o n .
* It is n e c e s s a r yt o sum s e v e r a l terms
of t h i s d i s t r i b u t i o n i n o r d e r t o o b t a i n t h e p r o b a b i l i t y t h a t a givensamplingplan

*Adescription of distributionsis given in the AppendixofVol. I11 - Testing

of t h i s series.

94
 Table 8-1
Categories of Testing Data and Associated Computational Problems
(Testing 1
Results

Attribute Data
Failure or nonf ailure
I

r Stress-Strength Measurement
Sensitivity Data
=
Go-no go

-Check consistency of observed -Test to failure data

data with requirements tensile strength
yield strength
-Provide sampling plans
-Perform sensitivity analyses
-Obtain confidence limit probit method
estimates based on binomial other
distribution
-Design experiment

Failure-time Data

I -Obtainform of life -Relate performance measure- -Perform the appropriate

distribution ments to component part time series analysis
parameters and environment autocorrelation
-Test for goodness of spectral densities
fit with assumed -Estimate unknown constants
distribution form in model; obtain estimates -Relate performance
of their precisions measurements to inputs,
-Estimate parameters parts, and environment
and characteristics -Use performance measure-
of distribution ments to screen parts -Check f o r consistency
of results with
-Check consistency of -Test adequacyof models requirements
observed data with relating system perform-
requirements ance to component and -Provide data sampling
stress parameters methods to yield
-Provide sampling plans required data for
(fixed sample size -Check consistency of computations
and sequential type) observed data with
requirements

95
 T a b l e 8-2
T e s t i n g - R e l a t e d Computer ProgramswithCorrespondingProblem Areas

1. A t t r i b u t e Data

A. L i b r a r y of
Programs
(1). BMD (Ref. 8-16)
(2) STAT-PACK (Ref. 8-7).

2. V a r i a b l e s Data

A. Performance Data ( a tD i s c r e t e Times)

(1) Least Squares
(Linear)
(a) STAT-PACK (Ref. 8-7)
(b) WHIRLPOOL (Ref. 8-2)
(c) IBM, 6.0.057
(Ref. 8-3)
(d) BMD programs(Ref. 8-16)
(2)Nonlinear Least Squares
(a) NOLLES (Ref. 8-1)
(b) SDA-3094
IBM-SHARE Library
(3)
General
Reference
(Ref. 8-1)
B. Performance Data (ContinuousRecords) - Autocovariance and Power Spectrum
(1) STAT-PACK (Ref. 8-7)
(2) BMD programs
(Ref. 8-16)
C. Failure-Time Data
( 1 )D i s t r i b u t i o nF r e e Estimates
Burn-inProcess,Estimation of HazardDataandLifetime,Density
Function(Ref. 8-15)
(2) Estimate of Parameters of Assumed D i s t r i b u t i o n s
(a)
Weibull (Ref. 8-13)
(b) Gamma (Ref. 8-8, 8-13, 8-14)
(c)Extreme Value (Ref. 8-11)
(d-) Log-Normal (Ref. 8-10)
(Le o) g i s t i c (Ref. 8-12)
( f ) Normal (Ref. 8-9)
Programscanbeobtained inconnectionwitheach of t h e a b o v e
a l t h o u g h t h e y may n o t b e s p e c i f i c a l l y i d e n t i f i e d i n t h e
references.

(3) Stress StrengthMeasurements -- S e n s i t i v i t y Data (Ref. 8-6)

 w i l l accept a l o t o f items g i v e n t h e t r u e p r o p o r t i o n d e f e c t i v e i n t h e l o t . Two
quality levels are chosen,onewhich is c o n s i d e r e d a c c e p t a b l e and t h e o t h e r c o n s i d e r e d
nonacceptable.These are r e f e r r e d t o as t h e a c c e p t a b l e q u a l i t y l e v e l (AQL) and t h e
l o tt o l e r a n c ep e r c e n td e f e c t i v e (LTPD) r e s p e c t i v e l y .
The p r o b a b i l i t y o f r e j e c t i n g a l o t of items g i v e n t h a t t h e q u a l i t y l e v e l is
equaltothe AQL i s c a l l e d t h e p r o d u c e r ' s r i s k . The p r o b a b i l i t y t h a t t h e l o t is
a c c e p t e dg i v e nt h a tt h ep r o p o r t i o nd e f e c t i v e is e q u a l t o t h e LTPD i s c a l l e d t h e
c o n s u m e r ' sr i s k .I f i t i s n e c e s s a r yt o compute t h e s e two r i s k s f o r a number of
problems, i t i s d e s i r a b l e t o h a v e a computerprogram toperformthenecessary
computations.
Many p r o g r a m s h a v e b e e n w r i t t e n f o r t h e s e p r o b l e m s , a n d t h e r e s u l t s h a v e b e e n
tablulated in a l a r g e number of t a b l e s o f s a m p l i n g p l a n s and by means ofgraphs
[Ref. 8-11. L i s t i n g s of t h e s e programshavenotbeenprovided i nt h el i t e r a t u r e ,
primarily because such programs are e a s y t o write.
Computations similar t ot h o s ed e s c r i b e da b o v e are n e c e s s a r y t o o b t a i n c o n f i d e n c e
limits f o r t h e t r u e p r o p o r t i o n o f f a i l u r e s p (ordefectives)in a l o t of s u b m i t t e d
items. I t is o f t e nd e s i r e dt oo b t a i na nu p p e r limt p and t o do s or e q u i r e st h e
U
s o l u t i o n of a n e q u a t i o n of t h e form:

X
0

where
x i s t ho
ebserved number o f a i l u r e s ,
0

is theupperconfidence limit,
PU
n is t h e number
of items i nt h es a m p l e ,
l-a i s t h ec o n f i d e n c el e v e l , and
a is t h er i s k of n o ti n c l u d i n gt h et r u ep r o p o r t i o no fd e f e c t i v e s
i nt h ec o n f i d e n c ei n t e r v a l 0 < p < p f o r p t h et r u ep r o p o r t i o n
U
of f a i l u r e s .
This equation can be solved by a n i t e r a t i o n p r o c e d u r e u s i n g t h e i n c o m p l e t e Beta
f u n c t i o n ofone of t h e t r a n s f o r m e d d i s t r i b u t i o n s s u c h as t h e v a r i a n c e r a t i o of an F
distribution. A programcan e a s i l yb ew r i t t e nt op e r f o r mt h er e q u i r e dc o m p u t a t i o n
i f o n es u p p l i e s as i n p u t s t h e n e c e s s a r y v a l u e s of t h e F d i s t r i b u t i o n o r i f onepro-
videsanapproximatingfunctiontothe F distributionforeachpossiblecombination
of i t s two parameters.The l a t t e r procedure would r e q u i r e c o n s i d e r a b l e i n p u t so a
simplerprocedurewouldbetosolvetheequation by a d i r e c t i t e r a t i o n p r o c e d u r e .

97

I
8.2 V a r i a b l e s Data
It i s c o n v e n i e n t i n t h i s s e c t i o n t o d i v i d e t h e v a r i a b l e s d a t a . i n t o t h e t h r e e
categories:
( 1f)a i l u r e - t i mde a t a ,
(2) performancemeasurements a t d i s c r e t et i m e ( s ) , and
(3) continuousrecordingofperformancemeasurements.
Thisclassificationofvariablesdata is made p r i m a r i l y f o r t h e c o n v e n i e n c e o f t h e
computationalprocedures;computerprogramsassociatedwiththeanalysesdonot
necessarilymatchtheclassificationof tests d e s c r i b e d i n V o l . I11 - T e s t i n go f
t h i s series. F o re x a m p l e ,t h e breakdownof performancemeasurements i n t ot h e two
c a t e g o r i e s ,d i s c r e t ev e r s u sc o n t i n u o u s ,c o r r e s p o n d st ot h ed i g i t a lv e r s u sa n a l o g
recordingmechanisms.Althoughboththesetypesofmeasurements are u s e d f o r t h e
same g e n e r a l p u r p o s e , t h e a n a l y t i c a l methods are q u i t e d i f f e r e n t .
8.2.1
Failure-time Data
I f a sampleof items are placedon t e s t f o r a f i x e d t e s t time o r u n t i l a
s p e c i f i e d number of f a i l u r e s h a s o c c u r r e d , t h e test r e s u l t s c o n s i s t of a s e t of
f a i l u r e times f o r t h e f a i l e d items and t h e t e r m i n a t e d test time f o r a l l items which
h a v en o tf a i l e d . I t is u s u a l l yd e s i r e d t o p r e d i c t , on t h eb a s i so ft h e s ed a t a ,t h e
behaviorof a l a r g ec o l l e c t i o no fi t e m st ob eu s e du n d e rs i m i l a rc o n d i t i o n s . When
p e r f o r m i n gt h i sp r e d i c t i o n ,c e r t a i np r o b l e m sm u s tb ec o n s i d e r e d :
( 1 )d i s c r i m i n a t eb e t w e e nt h ef o r m so ft h el i f ed i s t r i b u t i o n s ,e . g . ,n o r m a l ,
e x p o n e n t i a l ,W e i b u l l ,e t c . ,
(2) t e s t f o r goodnessof f i t w i t ha n assumed d i s t r i b u t i o nf o r m ,
(3) estimate t h ep a r a m e t e r so ft h ed i s t r i b u t i o n ,e . g . ,t h ef a i l u r e rate
parameter i n t h e case o f t h e e x p o n e n t i a l d i s t r i b u t i o n ,
( 4 )e s t i m a t ec h a r a c t e r i s t i c s of t h ef a i l u r er a t ed i s t r i b u t i o n ,
(5) p r o v i d et e s t i n gp l a n sa n dt h e i ra s s o c i a t e do p e r a t i n gc h a r a c t e r i s t i c s ,a n d
(6) c h e c kc o n s i s t e n c yo fo b s e r v e dd a t aw i t hc o n t r a c t u a lr e q u i r e m e n t s .
Some t e c h n i q u e s f o r d i s c r i m i n a t i n g b e t w e e n t h e f o r m s o f t h e l i f e d i s t r i b u t i o n s
havebeengiven intheliterature,for example, see R e f s . 8-4 and 8-5. Howeveryit
i s p o s s i b l e t o compute c r i t e r i a f o r g o o d n e s s - o f - f i t f o r e a c h o f t h e d i s t r i b u t i o n s
and select t h e p a r t i c u l a r f o r mg i v i n gt h eb e s tv a l u e of t h i s measure. Some s t a t i s t i c a l
programs a r e a v a i l a b l e f o r p e r f o r m i n g a goodness-of-fit,namely:
(1) Kolomogorov-Smirnov t e s t s , and
n

(2) X L tests.
Computer programs f o r t h e s e tests a r e i n c l u d e d i n STAT-PACK [Ref. 8-71. This
p a c k a g eo fp r o g r a m sa p p e a r st ob et h em o s tc o m p r e h e n s i v ep a c k a g ea v a i l a b l e t o date.

98
The programs are written for small to medium size computers
(8K words) and they do
not require any nonstandard features. They are card input and card and/or printer
I1 language. The
output oriented. The programs are written entirely in the FORTRAN
output of each program is lalieled as completely as possible for ease of understanding
by users.
One of the basic problems in comparing distributions is estimating the parameters
of each proposed distribution. Several programs are available for estimating the
parameters of the normal, log-normal, Weibull, gamma, generalized gamma, exponential,
extreme value, and logistic distribution. In particular, an entire series of FORTRAN
computer programs for this purpose are available upon request from the Aerospace
Research Laboratory (ARL), Wright-Patterson Air Force Base, Ohio. In addition to
these programs there is a collection of references describing the parameter estimation
procedures for each of the above distributions. Almost all of these are availabLe
in the published literature [Refs.8-9 through 8-14]. The estimation procedures are
iterative and based on the maximum likelihood method of estimation. Four programs
are included in STAT-PACK for estimating the parameters' of the normal, log-normal,
and the generalized gamma distributions. Some of the above programs include pro-
cedures for estimating the precision of the estimated parameters.
If one is unable to assume a particular form of the distribution, it may be
possible to make an assumption concerning the monotonic behavior of the hazard rate.
For example, this rate may decrease with time for many electronic components. In
such cases it is desirable to estimate the hazard rate at the end of the
A test.
paper appeared recently [Ref.8-151 on this subject and included the listing of a
program for obtaining confidence limits for the estimated failure rate at the termina-
tion of the test under the assumption
of decreasing failure rate.
A great many sampling plans have been provided in the literature under the assump-
tion that the failure time distribution takes on
of one
the many forms given above.
The program is not normally listed in connection with the computations of the
sampling plans; however, it is possible to write these programs in most
by cases
studying the discussions accompanying the tabulated results.
8.2.2 Performance Measurements at Discrete Time(s)
In this section performance measures such as the output or
voltage
the current
gain of an electronic circuit or the "hot spot" temperature in a nuclear reactor
core will be considered.
It is assumed that one wishes to relate these performance
measurements to characteristics of the component parts and the environmental stresses.
Very oftenit is possible to write these relatianships
on the basis of technical
knowledge concerning the circuit. On the other hand, it is sometimes possible
. only

99
 to relate the performance to certain part characteristics and environmental stresse
by means ofan analytical expressionin which certain constants or parameters are
, unknown but which can of an experiment. Some of the
be estimated from the results
problems which are typical are:
(1) to estimate unknown constants in the analytical models and obtain estimates
of the precisions of the constants and of the complete model,
(2) to use analytical models to screen out the "bad" components,
( 3 ) .to check the consistency of the observed data with the contractual
requirements, and
(4) to select the parts and their associated characteristics to optimize the
performance of a circuit.
To estimate the unknown constants in the analytical models, one can make use
If a model
any one of many computer programs based on the method of least squares.
is linear in the unknown constants to be estimated there are three basic approach
which have been programmed:
(1) fitting the complete model,
(2) fitting the model by adding on terms one at a time, called step-wise
regression, and
(3) fitting all combinations of linear models taking the variables one at a
time, two at a time, etc.
Several programs are included STAT-PACK
in for the approaches(1) and (2)
given above. Two programs are available for the third approach [Ref. 8-2,
8-31.
In case the model
is nonlinear in the constants to be estimated the least squares
procedure is still applicable, but the method of solution is iterative and based
on one of many possible searching techniques. Several programs have been written
or nonlinear regression problems [Refs. 8-1, 8-71.
In addition to the above men-
tioned programs one will find comparable programs in the CO-OP, and other
SHARE,
of the constants
such computer service systems.In order to estimate the precisions
certain additional computations must be performed, such as obtaining ofthe sum
squares of deviations of the observations from the predicted mean performance values
and inverting matrices. Most of the programs described above include some of these
additional computational features.
One technique used to screen bad components is to obtain a linear discriminati
function with the characteristics of the components.
The coefficients in the linear
function are estimated
'by an approach sinilar to that used in least square problems.
A computer STAT-PACK [Ref. 8-71 is available for performiag this analysis.
program in
Having determined the functional relationship an item is declared good only if, fo

100
I

example, theva.l.ue ofthe functionis less than or equal to a particular constant

c and is declared bad if the value of the function exceeds c. The
of val.ue
the
function is determined by substituting the characteristics of the components into
the discriminating linear relationship with known constants estimated from the data.
In order to select the component part of a system such that the performance will
be optimized, it
is necessary to obtain an analytical model relating performance
measures to the pertinent part characteristic and environmental stresses. Obtaining
this model has been previously discussed; it is assumed that such
a model has been
obtained from theoretical methods and/or experimental results. Given the model,
the problem then is to find the maximum or minimum value of the function for the
region of possible values of the part characteristics.
The many optimization programs
that are available or solving these problems were tabulated and discussed briefly
in the section on optimization techniques. Those techniques which would be of parti-
cular value here are the search techniques and nonlinear programming methods, because
it is expected that most of the relationshipsbe will
nonlinear. The optimization
techniques will yield the optimum values of the part characteristics from which one
can hopefully select the best parts to use in the system.
8.2.3 Continuous Recording of Performance Measurements
In order to assess the performance of many physical systems it is often necessary
.to record measurements continuously
by means of analog equipment. Although the use
of an purpose for taking such measurements does not differ from those taken at dis-
crete times, the analysis techniques are quite different. Hence this type of measure-
ment is treated separately. Typical computation problems that arise in this connection
are :
(1) performing time series analysis, including autocorrelation and spectral
density analyses
;
(2) relating characteristics of performance measurements to input, parts,
environmental characteristics;and
(3) providing data record sampling methods to yield the desired results and
the required degree of precision.
The usual procedure in analyzing continuous records is to select an appropriate
set of data at equal time intervals from the data tape of interest. These data sets
make up a time series which then become input to a standard computer program which
performs the autocorrelation and spectral density analysis. Many programs are
available to perform these computations; for example, STAT-PACK includes a time
The BMD package of statistical
series analysis and a time series plotter program.
programs [Ref. 8-16] contains two applicable programs; one performs a cross-spectral.

101
a n a l y s i s and t h eo t h e rp e r f o r m sr e l a t e dc o m p u t a t i o n s .S i m i l a rp r o g r a m s are a v a i l a b l e
throughcomputerserviceorganizations.
8.3 Stress-Strength'Measurements
A g r e a t many t e s t i n g p r o b l e m s f a l l into the category of d e t e r m i n i n g t h e s t r e n g t h
of t h e components t o b e u s e d i n a s y s t e m .A l t h o u g hs t r e n g t h may b e c o n s i d e r e d t o b e
a performancemeasurement in the general sense, i t is t r e a t e d h e r e i n a separate
sectionbecauseofthenature of t h e t e s t i n g p r o b l e m and t h e r e s u l t i n g d a t a .
It i s n o t a l w a y s p o s s i b l e t o p l a c e a n item on t e s t a n d i n c r e a s e t h e s t r e n g t h
i n a c o n t i n u o u s manner u n t i l t h e item f a i l s a n d u s e t h e stress a t t h e t i m e of f a i l u r e
as t h e s t r e n g t h o f t h e item. I n t e s t i n g many components t h e p r o c e d u r e is t o p l a c e
severalitems on t e s t a t e a c h ofseveral stress l e v e l s andobservethe number of
f a i l u r e s a t each stress l e v e l . From t h e s e t e s t r e s u l t so n ec a nd e r i v e a distribution
f u n c t i o nf o rt h ep r o b a b i l i t yo ff a i l u r ev e r s u st h e stress l e v e l . Such t e s t i n g i s
frequentlyreferredtointheliterature as s e n s i t i v i t y t e s t i n g . A largevariety
ofsensitivity tests h a v e b e e n d i s c u s s e d i n t h e l i t e r a t u r e ; t h e s e h a v e b e e n c o n v e n i e n t l y
summarized i n [Ref. 8-61. One of t h e e a r l i e s t s e n s i t i v i t y tests i s a s e q u e n t i a l
p r o c e d u r er e f e r r e dt o as t h eB r u c e t o no rt h e "up and down'' t e s t method. I nt h i st y p e
o fe x p e r i m e n tt h e iterms are t e s t e d o n e a t a time a t a stress l e v e l ; e a c h item t e s t e d
i s dependent on t h er e s p o n s ea n dt h e stress l e v e l o ft h ep r e v i o u s item t e s t e d . Many
variationsofthese tests havebeensuggested, most ofwhich are j u s t d i f f e r e n t pro-
c e d u r e s by whichonedeterminesthe stress l e v e l f o r e a c h item t e s t e d i n terms of
thelevelsusedfor a l l p r e v i o u s tests r a t h e r t h a n j u s t t h e l a s t - t e s t e d item.
Theanalysesofthedataresultingfromsuchexperiments are u s u a l l y q u i t e e a s i l y
performed by manualmethods.Consequently,only a few programs a r e a v a i l a b l e f o r
p e r f o r m i n gt h ea n a l y s e so f test d a t a r e s u l t i n g from s e n s i t i v i t ye x p e r i m e n t s .I n
p a r t i c u l a r , a program f o r a
*
p r o b i ta n a l y s i s is i n c l u d e d i n t h e BMD series ofprograms
[Ref.8-16]andone i n [Ref. 8-61,The l a t t e r r e f e r e n c ei n c l u d e si na d d i t i o n computer
programs f o r Monte C a r l o s i m u l a t i a n o f t h e t e s t r e s u l t s and t h e a n a l y s i s o f p r o p o r t i o n s
offailures by t h e method o f r e v e r s a l s . T h i s l a t t e r method i s f r e q u e n t l y u s e d i n t h e
a n a l y s i so fe x p e r i m e n t si nw h i c ht h e stress l e v e l i s determinedonthebasis of t h e
p r o p o r t i o n of s u c c e s s e so b s e r v e d a t a l l p r e v i o u s stress l e v e l s t e s t e d , and t h ep r o -
p o r t i o n of f a i l u r e s is assumed t o b e e i t h e r a n i n c r e a s i n g o r a d e c r e a s i n gf u n c t i o n
of t h e stress l e v e l .

*The
p r o b i t method i s a n o n s e q u e n t i a l d e s i g n f o r r e l a t i n g r e s p o n s e t o stress
orstimuluslevel.

102
 References
8-1. Nelson, A. C.; et. al.: Evaluation of Computer Programs for System Performance
Effectiveness. Progress Report No. 1 (Lab Project 920-72-1, SF-013-14-03,
Task 1604.,.Contract NO0140 66C 0499), Research Triangle Institute, System
Statistics Group.

8-2. Krumbein, W. C.; et. al.: Whirlpool, A Computer Program for Sorting
Out
Independent Variables by Sequential Multiple Linear Regression. Northwestern
1964, AD 611 142.
University, Evanston, Illinois,

8-3. IBM, 6.0.057: Linear Regression Analysis of All Combinations

of Variables.

8-4. Cox, D. R.: Further Results on Tests of Separate Families of Hypotheses.

University of London, J. Roy. Stat. SOC, (B), no. 24, 1962,pp. 406-424.

8-5. Cox, D. R.: Tests of Separate Families of Hypotheses. Proceedings of the

4th Berkeley Symp.,vol. 1, pp. 105-123.

8-6. Rothman, D.; Alexander,M. J.; and Zimmerman,J. M. : The Design and Analysis
of Sensitivity Experiments. NASA CR-62026, vols. I and11, May 1965.

8-7. Shannon, Stan; and Henschke, Claudia; STAT-PACK: A Biostatistical Programming

Package. Wadley Research Institute, Dallas, Texas, Commun. ACM, vol.
10,
no. 2, Feb. 1967, pp. 123-125.

8-8. Harter, H. Leon: Maximum-Likelihood Estimation of the Parameters of a Four-

Parameter Generalized Gamma Population From Complete and Censored Samples.
Applied Mathematics Research Laboratory,
ARL 67-0089, April 1967.

8-9. Harter, H. Leon; and Moore, Albert H.: Iterative Maximum-Likelihood Estimation
of the Parameters of Normal Populations from Singly and Doubly Censored
Samples. Biometrika, vol.53, nos. 1 and 2, 1966,pp. 205-213.

8-10. Harter, H. Leon; and Moore, Albert H.: Local-Maximum-Likelihood Estimation

of the Parameters of Three-Parameter Log-normal Populations from Complete
and Censored Samples. J. Am. Stat. ASSOC., vol. 61, September 1966,
pp. 842-851.

8-11. Harter, H. Leon;and Moore, Albert H.: Maximum-Likelihood Estimation, from

Doubly Censored Samples,of the Parameters of the First Asymptotic Distri-
bution of Extreme Values. Aerospace Research Labs and Air Force Institute
of Technology, Wright-Patterson AFB, Ohio.

8-12. Harter, H. Leon; and Moore, Albert

H.: Maximum-Likelihood Estimation, from
Censored Samples, of the Parameters
of a Logistic Distribution. Aerospace
Research Labs and Air Force Institute of Technology, Wright-Patterson
AFB,
Ohio.

8-13. Harter, H. Leon; and Moore, Albert H.: Maximum-Likelihood Estimation of the
Parameters of Gamma and Weibull Populations From Complete and From Censored
Samples. Technometrics, vol.7, no. 4, November 1965.

8-14. Harter, H. Leon: Series Expansions for the Incomplete Gamma Function and Its
Derivatives. Blanch Anniversary Volume,Aerospace Research Labs, Office
of Aerospace Research,U. S . Air Force,February 1967.

103

I
References (Continued)
8-15. Barlow, R. E.; Proschan, Frank;Scheuer, Ernest
M., with an Appendix by
Madansky, Albert: Statistical Estimation Procedures for the "Burn-In"
Process. R"5109-NASA, RAND Corporation, September 1966.

8-16. Dixon, W. J., ed.: Biomedical Computer Programs (BMD). Health Sciences
Computing Facility, Dept. of Preventive Medicine and Public Health, School
Los Angeles, Calif., Jan.1, 1964.
of Medicine, Univ. of Calif.,

104
9. Trends i nD i g i t a l Computation
In previous sections of the report, we have identified and d i s c u s s e d t h e v a r i o u s
a s p e c t s of d e s i g n f o r r e l i a b i l i t y w h e r e t h e computercanprovide assistance. In this
s e c t i o n w e summarize some recent developments in communicating with the computer which
promises t o make i t of much g r e a t e r v a l u e t o t h e s c i e n t i s t and t h ee n g i n e e r .T h e s e
developments are n o t s p e c i f i c a l l y r e l a t e d t o r e l i a b i l i t y , b u t s i n c e t h e y are of a
general nature their impact w i l l certainly be felt in many f u t u r e u s e s o f t h e c o m p u t e r
forreliabilityanalyses.
There are threecomputerdevelopments we wishtodiscuss.First,theuse of
problem-orientedlanguages is c e r t a i n t o s p r e a d as t h e y c o n t i n u e t o b e d e v e l o p e d and
their utility becomes b o t h g r e a t e r andmorewidelyappreciated. The o t h e r two develop-
ments are on-linecomputation a n dc o m p u t e rg r a p h i ci n p u t l o u t p u tc a p a b i l i t i e s ,h e r e
simply called graphics.
Problem-orientedlanguages are a l r e a d yi nw i d eu s e[ R e f s . 9-1 t o 9-41. A problem-
orientedlanguagepermitsthedescription of a broad class ofproblems i n a given
problem area v i a a simple vocabulary comprised of terms f a m i l i a r t o t h e e n g i n e e r w o r k i n g
inthat problem area. F o re x a m p l e ,t h ee l e c t r o n i cc i r c u i ta n a l y s i sp r o g r a m ECAP i n p u t
languageusesforthemostpartthe same n o m e n c l a t u r e t o d e s c r i b e a circuit to be
analyzedthatthecircuitanalysisengineer would u s e t o a n a l y z e t h e c i r c u i t by hand.
A computerprogram w r i t t e n i n a problem-oriented language i s n o t a program i n t h e
o r d i n a r ys e n s e .T h i s i s because i t is r e a l l y j u s t a n unambiguousproblem description
ratherthanthelogicalsequenceofstepsrequiredforthesolution of theproblem.
The s e q u e n c e( i . e . ,t h ea l g o r i t h m )r e q u i r e dt oi m p l e m e n tt h es o l u t i o n of a given
problem is i n c o r p o r a t e d as a p a r t of t h e computerprogram forprocessinginput state-
ments t ot h ep r o b l e m - o r i e n t e dl a n g u a g e ;t h e s ei n p u ts t a t e m e n t s a r e theproblemdescrip-
tion. Thus t h e programmer o rt h ed e s i g n e rd o e sn o tn e e dt ow o r r ya b o u tw h e t h e rh i s
algorithmforsolvingtheproblem is c o r r e c t ; h e n e e d o n l y w o r r y t h a t h i s p r o b l e m is
p r o p e r l y andunambiguously stated.Problem-orientedlanguageshavealreadybeen
d e v e l o p e df o ru s ei nd e s i g n i n gc h e m i c a lp r o c e s s i n gp l a n t s[ R e f . 9-11, structures
[Ref. 9-21, and e l e c t r i c a lc i r c u i t s[ R e f s . 9-1, 9-3,and 9-41.
The o n l y r e s p o n s i b i l i t y o f t h e p r o b l e m - o r i e n t e d l a n g u a g e u s e r is t h a t h e know
t h es y n t a xo ft h ep r o b l e m - o r i e n t e dl a n g u a g e and some s i m p l e r u l e s c o n c e r n i n g t h e o r d e r i n g
of t h es t a t e m e n t sw h i c hd e s c r i b et h ep r o b l e mh e i s solving. To summarize, t h e problem-
o r i e n t e dl a n g u a g e i s simply a s p e c i a l programwhichallows as i n p u t t h e unambiguous
d e s c r i p t i o n of a p a r t i c u l a r p r o b l e m s u i t e d t o t h a t l a n g u a g e a n d t h e a s s o c i a t e d d a t a
r e q u i r e df o rs o l u t i o no ft h eg i v e np r o b l e m . The i n d i v i d u a l u s i n g t h e l a n g u a g e writes
a new i n p u t p r o g r a m f o r e a c h d i f f e r e n t p r o b l e m , w i t h o u t h a v i n g t o w o r r y a b o u t t h e

105
problemsolutionalgorithmincontrasttoprocedure-orientedandassemblylanguage
programs. To p r o v i d et h ea d v a n t a g e so ft h ep r o b l e m - o r i e n t e dl a n g u a g et h e r em u s t
be a c o m p u t e r p r o g r a m w h i c h p r o c e s s e s t h e i n p u t s t a t e m e n t s , d i g e s t s t h e i n f o r m a t i o n
contained in the statements, and generates the proper machine language program which
when e x e c u t e d s o l v e s t h e p r o b l e m d e s c r i b e d by t h e i n p u t s t a t e m e n t s .
O n - l i n ec o m p u t a t i o nr e f e r st ot h es i t u a t i o nw h e r e i nt h ec o m p u t e ru s e r sits a t
t h e computerconsole ( i t may b e t h e c o n s o l e of a small computer j u s t f o r t h e s i n g l e
programmer, o r i t may be a r e m o t e t e r m i n a l c o n n e c t e d t o a l a r g e c e n t r a l computer)
andviews theresults of h i s p r o g r a mi n s t a n t a n e o u s l y .I nt h ee a r l yd a y s of d i g i t a l
computation, i t was a p r a c t i c a l t h i n g f o r t h e d e s i g n e r o r o t h e r c o m p u t e r u s e r s t o u s e
t h e computer i n t h i s f a s h i o n . However, as themachines became b i g g e r , more powerful,
and more e x p e n s i v e , i t became n o l o n g e r p r a c t i c a l f o r t h e i n d i v i d u a l t o u s e t h e c o m p u t e r
i nt h i si n d i v i d u a lf a s h i o n . The r e s u l t w a s t h a t computermonitoringprogramscalled
o p e r a t i n gs y s t e m s were d e v e l o p e dt os u p e r v i s et h eo p e r a t i o no ft h em a c h i n e . Computer
systemsoperatinginthis mode p r o c e s s t h e c o m p u t e r u s e r ' s p r o g r a m i n a sequential
f a s h i o n , s o thateachprogram i s c o m p l e t e l yf i n i s h e db e f o r et h en e x tp r o g r a m i s begun.
The mode is c a l l e db a t c hp r o c e s s i n g . Thecomputer f a c i l i t y is o p e r a t e di ns u c hc a s e s
(and t h i s i s t h eu s , u a lc a s et o d a y ) on a closed-shopbasis,which means t h a tt h ec o m p u t e r
u s e r is n o t p r e s e n t when h i s program i s b e i n g r u n and t h e time d e l a y b e t w e e n d e l i v e r y
oftheprogram t o berunandthereturnofthecomputerresults(theso-calledturn-
a r o u n dt i m e )v a r i e sf r o mh o u r st od a y s .
Becauseof theturn-aroundproblem, i t simply is n o t p r a c t i c a l t o u s e t h e c l o s e d -
shopcomputer t o s o l v e problems by h e u r i s t i c m e t h o d s , e x t r a p o l a t i n g earlier successes
t o o b t a i n new ones,Because of t h e n a t u r e o f e n g i n e e r i n g d e s i g n , many ofthemost
c h a l l e n g i n ge n g i n e e r i n gp r o b l e m s are most e f f e c t i v e l y s o l v e d by suchmethods.If
t h e computer is t o b e o f maximum a s s i s t a n c e i n t h i s d e s i g n p r o c e s s , t u r n - a r o u n d times
of h o u r s are o b v i o u s l yh o p e l e s s l yl o n g . Even turn-around times of minutes are u s u a l l y
too long to allow the designer to use the heuristic method of solving problems while
interactingwiththecomputer.
The p r o v i s i o n o f a method f o r a l l o w i n g t h e d e s i g n e r t o i n t e r a c t d i r e c t l y w i t h
t h e computercanbeobtainedeither by t h e small i n d i v i d u a l c o m p u t e r o r by a . r e m o t e
c o n s o l el i n k e dt o a largecomputer.Althoughbothapproacheshave merit, t h er e m o t e
t e r m i n a ll i n k e dt ot h el a r g ec o m p u t e r i s perhaps more v a l u a b l e i n t h i s r o l e .
S i n c et h er e s p o n s e time o f t h e d e s i g n e r is q u i t e s l o w compared t o t h e c o m p u t e r ,
theinstanteousresponseofthecomputertotherequestofthedesignercanbepro-
videdeconomicallyonlyiftheresourcesofthecomputer are s h a r e d f o r o t h e r p u r p o s e s
w h i l et h ed e s i g n e r i s thinkingandmodifyinghisprograms, etc. I t a p p e a r sc e r t a i n

106
that the near future will see the widespread use of computer consoles by e n g i n e e r i n g
personnelforeffectivedesigner-computerinteraction,inthistime-shared mode of
computing.Ref. 9-1 c o n t a i n s some examplesofsuchuses.
Thedevelopmentofcomputer g r a p h i c s is b o t h a powerful additional computer capa-
b i l i t y i n i t s e l f and a complement t o t h e above-discussed new computerdevelopments.
Thedevelopmentof effectivegraphicinput-outputdevicesforcomputerstraditionally
haslaggedthedevelopment of computers,and it is only quite recently that versatile
g r a p h i ci n p u t - o u t p u td e v i c e sh a v e become a v a i l a b l e a t r e a s o n a b l e c o s t . T h e f i r s t
graphical output device was doubtless the line printer wherein a c l e v e r programmer used
a p p r o p r i a t e l yc h o s e nc h a r a c t e r st os k e t c h a g r a p ho r a c r u d ep i c t u r e . Then,program-
c o n t r o l l a b l ec a t h o d er a yt u b eo u t p u td e v i c e s became a v a i l a b l e .A l t h o u g ht h ee a r l y
ones were q u i t e l i m i t e d i n t h e i r c a p a b i l i t i e s , t h e y p r o v i d e d g r e a t improvementsover
l i n ep r i n t e r su s e dt op r o d u c ep i c t u r e s .C a t h o d er a yt u b e sw i t hg r a p h i c a li n p u tc a p a -
bilityinadditiontooutputfirst became a v a i l a b l e i n t h e e a r l y 1 9 6 0 ' s . Some s u c h
e q u i p m e n ta l l o w st h ed r a w i n go fl i n e sd i r e c t l y on t h e f a c e o f t h e s c o p e u s i n g l i g h t
pens o ro t h e ri n p u td e v i c e s .O t h e r sa l l o wo n l yt h ed i s p i a yo fc h a r a c t e r s at fixed
l o c a t i o n so n l y ;t h ec h a r a c t e r s are t y p i c a l l yi n p u tv i a a t y p e w r i t e r - l i k ei n p u td e v i c e .
A considerable variety ofimproved graphicaldevicesforcomputers are c u r r e n t l y
underdevelopment.Thesedevices when p o s s e s s i n g l i n e d r a w i n g c a p a b i l i t i e s r e q u i r e
quite high transfer rates between t h e d i s p l a y d e v i c e a n d t h e c o m p u t e r t o m a i n t a i n
p i c t u r ec l a r i t y .C o n s e q u e n t l y , i t is common t o f i n d a small computer whose s o l e j o b
i t is t o m a i n t a i n and m a n i p u l a t e t h e d i s p l a y i n f o r m a t i o n , c o n n e c t e d t o a l a r g e computer
which p e r f o r m s t h e c o m p u t a t i o n s r e q u i r e d f o r t h e p r o b l e m u n d e r s t u d y .
T y p i c a l of what can be done with the combination of graphic input/output devices
i n a largepowerfulcomputer i s t h e DAC system[Ref. 9-11 developed by t h eG e n e r a l
M o t o r sR e s e a r c hC e n t e r .T h i ss y s t e m ,i na d d i t i o nt op r o v i d i n gd i r e c tc o m m u n i c a t i o n
between t h e d e s i g n e r a n d a powerfulcomputer,canproducecontroltapesforautomatic
d r a f t i n gm a c h i n e s ,n u m e r i c a l l yc o n t r o l l e dm i l l i n gm a c h i n e s , etc. Such systemsappear
destinedtoplayimportantrolesinthedesign of a l l f u t u r e complex e n g i n e e r i n g s y s t e m s .
The combination of a l l t h r e e of t h e abovedevelopmentshasalreadybeen made on
a ne x p e r i m e n t a lb a s i s[ R e f s . 9-5 and 9-61. I nt h e s ee f f o r t sc i r c u i ta n a l y s i s programs
were u s e do n - l i n ev i ag r a p h i ci n p u t / o u t p u td e v i c e st ot h ec o m p u t e r .T h o s ep e o p l e
who have used these experimental systems are h i g h l y e n t h u s i a s t i c a b o u t t h e e f f e c t i v e
i n c r e a s ei nd e s i g nc a p a b i l i t yt h r o u g ht h eu s e of t h e s e s y s t e m s . C e r t a i n l y t h e f u t u r e
w i l l see s u c h s y s t e m s p l a y i n g a n i m p o r t a n t r o l e i n i m p l e m e n t i n g p r e s e n t l y a v a i l a b l e
and f u t u r e more g e n e r a l r e l i a b i l i t y a n a l y s e s .

107
References

9-1. Katz, D. L . ; e t . a l . : Computers i n EngineeringDesignEducation, Vol. I -

Summary. The U n i v e r s i t y of Michigan,College of E n g i n e e r i n g ,A p r i l 1, 1966.

9-2. Fenves, S. J . ; e t . a l . : STRESS-A User's Manual. MIT P r e s s , Cambridge, Mass.,

1964.

9-3. IBM P u b l i c a t i o n H20-0170-1. 1 6 2 0 . E l e c t r o n i cC i r c u i tA n a l y s i sP r o g r a m (ECAP)

(1620-EE-02X) User's Manual.Tech. P u b l i c a t i o n sD e p t . ,W h i t eP l a i n s , N. Y . ,
1965.

9-4. Computer-Aided CircuitDesignSeminarProceedings.KresgeAuditorium, MIT,

Cambridge, Mass., A p r i l 11-12, 1967, NASA TMX-59610.

9-5. S o , H. C.: OLCA - An O n - l i n eC i r c u i tA n a l y s i sS y s t e m P

. u b l i s h e di n Computer-
Aided CircuitDesignSeminarProceedings.KresgeAuditorium, MIT, Cambridge,
Mass., A p r i l 11-12,1967,pp. 9-11, NASA TMX-59610.

9-6. H o g s e t t , G. R.; Nisewanger, D. A.; and O'Hara, A. C . , Jr.: ApplicationExperiment

withOn-lineGraphics-Aided ECAF'. 1967 I n t e r n a t i o n a lS o l i d - s t a t eC i r c u i t s
ConferenceDigest of T e c h n i c a l P a p e r s , U n i v e r s i t y of P e n n . , P h i l a d e l p h i a ,
Pa.,Feb.1967,pp. 72-73.

108
 Appendix A

R e v i s e dV e r s i o n of PVA P r o g r a m L i s t e d i n Ref. 4-3

109
110
111
 24

25
26

27
29

29

30
31
C
C
C

32
33

112
113
55 FORMAT(l9X,
56 FORMAT(16F5.0)
57 F O R M A T ( 1 9 H O I N p UCr O R R E L A T I O N S / / )
59 F O R M A T ( 1 H, 2 0 F 5 . 3 )
59 FORHAT(1H-,SX,3(5X,A4,3X,)(5X.A4.3X)/8(5X,A4,3X))
6 0 FORMAT(lH0,5X~BE12.4/3X,BEl2.4/OE12.4)
6 1 F O R M A T ( 1 2 H - I N l ' l lCr H E C K )
62 F O R M A T ( l H - )
63 F O R M A T ( 4 1 H - D E p E N D E N TD A T AL I S T E DI NA S C E N D I N GO H D E R , / / 4 H
1 5 X , S H, I5/(N7 X s A 4 , 3 X ) )
6 4F O R H A T ( I 4 , F 1 0 . 35, E 1 4 . 4 )
65 F O R M A T ( 6 H - M O H E V 1 S / l O X 1 5 ( 7 X , A 4 . 4 X ) )
6 6 F O R M A T ( 1 FOIHK O ST,5El5.6)
6 7 F O R M A T ( 1 OSHEO COND,515.6)
68 FORHAT(1OHO 'IHIRD,5E15.6)
69 FORMAT(1OHO COUHTH,5E15.6)
7 0 F O R H A T ( ~ O H O S T DD. f V . , 5 E l 5 , 6 )
7 1 F O R H A T ( I O H OS K E w N E S S , 5 E l 5 . 6 )

73 F O R H A T ( 3 6 H O V A K I A N C E -
7 2 F O R H A T ( 1 O H OK U ? l O S I S , 5 E 1 5 . 6 )
C O V A R I A N C EM A T R I XO, R D E R e I 2 )
7 4 F O R H A T ( ~ H O , ~ X I P ~ , ~ X I ~ E ~ ~ , ~ )

114
115
 +A

l+X-DEN1)+94* (-X++3 +3.*(AHD+3.)aX*X-3.+

2(AMD+3.)*(AMD+2.)+X+DEN2)
ELPH(J,I)=l.+EXP(-X)+(TERMl+(X*+LAMDA)+TERH2)/COE
0 CONTINUE
RETURN
END

DO 1 I=l,N
RC=33.*XN+101.
XP=RC/2048.
MU=XP

116
 117

L
 ARG=O.
DO 1 I = l , N D F
CALL N O R M ( A h G 1 )
AR1; = ARG + AK:1 * AKGl
1 CONTINUE
ARG = AHG THEIA
RETURN
END

SOUHCE L I S T

118
 Appendix B

Bounds for Reliability Program

119
120
121
122

""
123
124
 Appendix C

R e l i a b i l i t y Cost Trade-Of f Analysis PrOgtam

125
 'I

c
C
c
C
c
C
C
L:
0
c
C
L:
c
C
c
C
I:
C

126
 127

I-
128
129
130
191